A transfer appropriate processing approach to investigating implicit memory for emotional words in the cerebral hemispheres

Neuropsychologia 43 (2005) 1529–1545

A transfer appropriate processing approach to investigating implicit memory for emotional words in the cerebral hemispheres Marjorie A. Collins ∗ , Amanda Cooke School of Psychology, Murdoch University, Murdoch, Perth, WA 6150, Australia Received 12 April 2004; received in revised form 1 October 2004; accepted 9 November 2004 Available online 16 February 2005

Abstract Forty undergraduate students participated in two experiments designed to investigate the impact of perceptual and conceptual encoding manipulations on implicit memory for emotional words in each cerebral hemisphere. Adopting a transfer appropriate processing approach, the encoding manipulations were designed to promote processing of the surface features of stimuli in Experiment 1, and their semantic meaning in Experiment 2. In both experiments, participants completed the designated encoding task, followed by a lexical decision task where primed and unprimed words were presented to the left (LVF) and right visual fields (RVF). In Experiment 1, implicit memory was observed for RVF presentations of words primed according to their perceptual features. Word valence did not impact on visual field of presentation for primed or unprimed words. In Experiment 2, participation in the conceptual encoding task differentially impacted on processing and implicit memory for emotional words presented in the LVF, where priming the conceptual meaning of words facilitated the processing of positive, relative to negative and non-emotional words. In addition, implicit memory for conceptually primed negative words was reflected in inhibition of primed relative to unprimed negatively valenced words presented in the LVF. In contrast, for RVF presentations, there was evidence of implicit memory for conceptually primed non-emotional words, but not for emotional words. The results are generally consistent with the right hemisphere model of emotion, which posits greater right hemisphere involvement in both the processing and implicit memory of emotional stimuli. The results also support Nagae and Moscovitch’s suggestion [Nagae, S., & Moscovitch, M. (2002). Cerebral hemispheric differences in memory of emotional and non-emotional words in normal individuals. Neuropsychologia, 40, 1601–1607] that level of processing be incorporated into studies examining the veracity of the right hemisphere and valence models of emotional processing. The study demonstrated the usefulness of adopting a transfer appropriate processing approach to investigating memory for word valence in each hemisphere. © 2005 Elsevier Ltd. All rights reserved. Keywords: Right hemisphere model; Valence model; Implicit memory; Emotional memory; Transfer appropriate processing; Perceptual processing; Conceptual processing; Cerebral hemispheres

1. Introduction Understanding the contribution of each cerebral hemisphere to memory of emotional material has the potential to elucidate the neural substrates and cognitive processes underpinning the formation and maintenance of cognitive biases in anxiety and depression. It is clear that the emotionality of a word influences both perception and memory of that word. Explicit recall is enhanced for emotionally valenced words relative to non-emotional words (Bock, 1986), recall ∗

Corresponding author. Tel.: +61 8 9360 2858. E-mail address: [email protected] (M.A. Collins).

0028-3932/$ – see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.neuropsychologia.2004.11.029

is superior for positively valenced stimuli relative to negative stimuli (Cacioppo, Petty, & Morris, 1985; Riskind & Lane, 1987) and implicit biases in processing emotional information are evident in both clinical and non-clinical groups (e.g. Denny & Hunt, 1992; Hocking & Collins, in press; Williams & McDowell, 2001; Williams, Watts, MacLeod, & Mathews, 1997). It is also clear that there are hemispheric differences in the processing of emotional words (e.g. Borod et al., 1998; Kinsbourne & Bemporad, 1984). However, little is known about the role each hemisphere plays in memory of emotional linguistic stimuli. The current study redresses this. Clinical and experimental research has cohered to produce two major neuropsychological models of hemispheric

1530

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

lateralization and emotional processing. The right hemisphere (RH) model proposes that this hemisphere is specialized for all emotional processing, irrespective of stimulus valence (Borod, 1992; Borod et al., 1998). The right hemisphere’s role in processing emotional information is thought to be associated with its substantial involvement in mechanisms of autonomic and behavioural arousal (Gainotti, Caltagirone, & Zoccolotti, 1993; Heller, 1993) with the integrative, gestalt processing style of the right hemisphere rendering it eminently suitable for the complex demands of emotional processing (Mandal, Tandon, & Asthana, 1991). In contrast, the valence model contends that emotional processing is differentially mediated by each hemisphere according to valence, with the right hemisphere specialized for negative emotion and the left hemisphere specialized to process positive emotion (Davidson, 1992; Davidson, Ekman, Saron, Senulis, & Friesen, 1990; Kinsbourne & Bemporad, 1984). This has been attributed to underlying differences in the neural representation of basic approach/withdrawal behaviours in each hemisphere (Davidson, 1992, 1998). Most support for the RH model stems from research using patient populations (e.g. Borod et al., 1992; Cicone, Wapner, & Gardner, 1980). While some of the studies on hemispheric processing of emotional words in neurologically normal individuals support the RH model (Brody, Goodman, Halm, Krinzman, & Sebrechts, 1987; Graves, Landis, & Goodglass, 1981; Nagae & Moscovitch, 2002; Richards, French, & Dowd, 1995; Schmitt, Hartje, & Willmes, 1997), some support the valence model (Ali & Cimino, 1998; Coney & Fitzgerald, 2000; Van Strien & Morpurgo, 1992), and others have failed to support either model (Eviatar & Zaidel, 1991; Leventhal, 1988; Strauss, 1983). Nagae and Moscovitch (2002) suggest that these inconsistencies are attributable to the type of task used to examine emotional processing. They argue that tasks involving word identification obscure the right hemisphere’s contribution to emotional processing, as the left hemisphere plays a central role in the early perceptual processing of words. They also argue that hemispheric differences in emotional processes are likely to manifest at later stages of processing, when analysis of higher-order attributes, such as emotional meaning, takes place. Nagae and Moscovitch supported this argument in a study comparing hemispheric contributions to emotional words incorporated into perceptual identification and explicit recall tasks. When emotional and non-emotional words were incorporated into a perceptual identification task, word emotionality did not interact with visual field of presentation. In contrast, hemispheric differences were evident in a task involving explicit recall of emotional words. They concluded that explicit memory for emotional words was more dependent on the right hemisphere, while perceptual identification of words was more dependent on the left hemisphere. This raises the possibility that hemispheric differences in emotional processing will emerge in tasks reflecting memory for emotional linguistic material, rather than tasks primarily tapping perceptual processing. Ali and Cimino (1997) have

provided evidence consistent with this proposal. They asked students to complete a lexical decision task presented in a divided visual field format and incorporating emotional and non-emotional words, followed by word recall and delayed word recognition tasks. Their findings were consistent with the valence model, with left hemisphere mediation of explicit memory for positively valenced emotional words and right hemisphere involvement in explicit memory of negative words. Ali and Cimino (1998) found further support for the valence model in a more extensive study examining lateralization for both implicit and explicit memory of emotional words. After completion of a lexical decision task, where emotional and non-emotional words were presented to each visual field, half of the participants completed explicit memory tasks (free recall and recognition), while the remainder completed an implicit memory task (word stem completion). Implicit and explicit memory of emotional valence differed in each hemisphere. In the left hemisphere, implicit memory was superior for positive words while explicit memory was superior for both positive and non-emotional words. In the right hemisphere, implicit memory was superior for negative and non-emotional words, while explicit recognition was superior for negative words alone. Hence, it appears that a fruitful means of clarifying the nature of emotional processing in each hemisphere is to examine memory for emotional information. In doing so, it is necessary to disentangle implicit and explicit memory. A viable approach is to adopt the framework provided by the transfer appropriate processing (TAP) model of memory, which is based on the notion that the cognitive operations employed during encoding have a profound impact upon subsequent recall (Roediger & Blaxton, 1987). Within a TAP framework, a distinction is drawn between perceptual and conceptual processing (Roediger & McDermott, 1993; Roediger, Weldon, & Challis, 1989). Perceptual processing involves processing the surface/physical features of stimuli, largely in the absence of semantic analysis, while conceptual processing requires interaction with the test material (e.g. elaboration, reconstruction and organization) to access its semantic meaning (Leshner & Coyle, 2000). Recall is a function of recapitulation of processes at encoding and retrieval. Most explicit memory tests involve conceptual processing, while most implicit memory tests involve perceptual processing (Roediger et al., 1989). TAP procedures have been used successfully to investigate implicit and explicit memory for emotional information. For example, Watkins, Martin, and Stern (2000) investigated mood congruent memory bias for positive and negative words using TAP methodology. Clinically depressed and non-depressed participants were assigned to either a perceptual or conceptual encoding condition and then completed implicit memory tasks that were perceptually based (i.e. word stem completion and word identification) or conceptually based (i.e. free association and word retrieval). Mood congruent memory bias was evident in the conceptually based word retrieval task, but only if stimuli had been conceptually encoded. Evidently, conceptual encoding enhanced implicit

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

memory for the emotional meaning of the words, whereas perceptual encoding did not. Similarly, Hocking and Collins (in press) adopted a TAP approach to compare implicit and explicit memory biases in students with high and low speech anxiety. We found an explicit memory bias for speech-related words in participants with high speech anxiety, after encoding which encouraged processing of the conceptual meaning of stimuli, but not after encoding which encouraged processing of the perceptual features of stimuli. The present study aims to investigate implicit memory in each cerebral hemisphere for emotional words using a new approach. In doing so, the veracity of the RH and valence models of emotional processing will be assessed. By adopting a TAP approach to determine the impact of perceptual and conceptual encoding manipulations on implicit memory for emotional words in each hemisphere, we will concurrently assess Nagae and Moscovitch’s (2002) argument that valence effects are more readily detected in the later stages of processing when semantic meaning is analysed, with the perceptual processes involved in word recognition a confounding factor in studies of this nature. To achieve this, two encoding conditions will be incorporated into the design, with participants required to encode emotional and non-emotional words according to their surface features in a perceptual processing condition (in Experiment 1), or according to their semantic meaning in a conceptual processing condition (in Experiment 2). Perceptual and conceptual processing will be operationally distinguished in accordance with Jacoby’s (1983) read/generate manipulation, which has been used to derive perceptual and conceptual tasks involving implicit memory for verbal material (Roediger et al., 1989). Following the encoding exercises, participants will complete a lateralized lexical decision task incorporating words from the encoding phase along with new words. This will provide a measure of implicit memory for previously presented words.

2. Experiment 1 The first experiment will assess the impact of perceptual encoding upon the formation of implicit memory for emotionally valenced words in each cerebral hemisphere. As counting the number of vowels or letters in words reliably engages perceptual processes (Leshner & Coyle, 2000; Mulligan, Guyer, & Belland, 1999) we will use a perceptual encoding task where participants count the number of long straight lined strokes in target words printed on a page. This is expected to encourage participants to focus on encoding the surface/physical features of these words. Following this encoding task, participants will complete a lexical decision task where they discriminate between word and non-word letter strings presented in the LVF and RVF. Half of these words will appear beforehand in the perceptual encoding task (henceforth referred to as primed stimuli) and half will be new words (henceforth called unprimed stimuli).

1531

Nagae and Moscovitch’s (2002) supposition that hemispheric differences in emotional processing will be evident in tasks evoking semantic analysis and not tasks primarily tapping perceptual processes, predicts that there will be no hemispheric differences in responses to emotional words which have previously been perceptually encoded. In contrast, a LVF advantage for emotional words relative to non-emotional words would support the RH model. A RVF advantage for positive words and a LVF advantage for negative words would support the valence model. 2.1. Method 2.1.1. Participants Forty undergraduate psychology students participated to satisfy course requirements. All gave informed consent prior to data collection, which was undertaken in accordance with approved ethical standards. Data for 11 participants were discarded as they had error rates exceeding 30% in either the word or non-word condition. The final sample comprised 17 female and 12 male participants ranging in age from 17 to 49 years, their mean age being 23.48 years. All participants reported normal or corrected-to-normal vision and spoke English as their first language. All were predominantly right handed as determined by Bryden’s (1977) simplified handedness questionnaire. The mean handedness laterality quotient for the final sample was +0.93. 2.1.2. Apparatus Participants were tested in a quiet, well-lit cubicle. Stimuli for the lexical decision task were presented on a Pentium personal computer with a software program which controlled randomization of stimulus presentation, trial sequencing, timing and data collection. These stimuli were presented in large yellow uppercase letters on a dark grey background, with each letter 10 mm in height and 7 mm in width, with 1 mm spacing, on a screen adjusted to the minimum brightness level to minimize after-effects due to phosphor persistence. All stimuli were presented with their innermost boundary appearing 2 degrees of visual angle to the left or right of a central fixation and subtending a horizontal visual angle between 2 and 7 degrees when participants were 60 cm from the VDU. Eye movements were monitored by a video camera connected to a monitor viewed by the examiner. An adjustable chin rest was used to ensure participants’ heads remained in the correct central position and distance from the VDU. 2.1.3. Design The principal dependent variable was reaction time (RT). Errors were treated as a subsidiary variable and used to check for speed-accuracy trade-offs. Three within subject variables were manipulated: priming condition, word valence and visual field of presentation. The only between-subjects variable was the order in which the positive and negative word blocks were presented. The first independent variable

1532

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

comprised primed and unprimed words. RT to 120 primed words provided the basis for making inferences on implicit memory for the perceptual aspects of emotionally valenced words, while RT to 120 unprimed words provided a measure of general processing of emotional words. The word valence variable comprised positive, negative and non-emotional words. Positive and negative words were presented in separate blocks, along with an equal number of non-emotional words which were randomly intermixed with the emotional words. The non-emotional words provided a baseline comparison for hemispheric contributions to processing and implicit memory for emotional words. All stimuli were matched on word length, imagery and frequency (refer to Section 2.1.4). Presenting positive and negative words in separate blocks had the advantage of decreasing the number of stimuli required for the encoding task, and reducing the number of consecutive blocks of trials in the lexical decision task. We anticipated that this would improve the quality of data by allowing participants to focus attention on a smaller sub-set of stimuli in the encoding task, and reduce their fatigue and boredom in the lexical decision task. It also had the advantage of encouraging participants to develop a ‘positive’ or ‘negative’ set bias, which we expected to maximize valence effects in a manner consistent with studies manipulating mood through repeated exposure to positively or negatively valenced statements (e.g. the Velten procedure, see Williams et al., 1997). Positive and negative word blocks were presented in counterbalanced order. The third experimental variable was visual field, where each target in the lexical decision task was presented to either the LVF or RVF. Selection of visual field, and order of presentation of stimuli within each word block, was randomly determined by the software program. Thus, for the lexical decision task, each word was presented once only, to either the left or right visual field, with each participant exposed to a unique stimulus sequence in each visual field. 2.1.4. Materials Positive and negative word sets were constructed to assess hemispheric contributions to the processing of, and implicit memory for, word valence.1 Each set comprised 60 emotional words: either 60 positive (e.g. hug, fun, passion) or 60 negative words (e.g. bad, prison, torture); and 60 non-emotional words (e.g. rod, mutton, cotton). All words were imageable nouns or adjectives. Each word set also included 90 orthographically legal and pronounceable non-words, which were derived by unsystematically changing between 1 and 3 letters of target words. Stimuli in the positive and negative word sets were equally divided, for use in the primed and unprimed word conditions. Half of the stimuli for each valence type were used in the primed word conditions, and so were presented in both the perceptual encoding and lexical decision 1 The stimulus materials are available at the following website: http://www.psychology.murdoch.edu.au/publications/collins stimlistcandc. html.

tasks. The remaining half of the stimuli were used in the unprimed word condition and so were presented only in the lexical decision task. So in the lexical decision task, the 30 words that had been perceptually primed were presented with 30 unprimed words, and 45 non-words. These words matched on all relevant word dimensions (see below). Stimuli were selected for inclusion in the positive and negative word sets if they met designated criteria for ratings on dimensions of word goodness, emotionality, imageability, and frequency. Ratings for goodness, emotionality and imageability were obtained from published norms based on a 7-point Likert scale (Brown & Ure, 1969; Paivio, Yuille, & Madigan, 1968; Rubin & Friendly, 1986; Summers, 1996; Toglia & Battig, 1978). Goodness reflects how intensely good (positive), or bad (negative) the meaning of the word was rated. Words selected for the positive set had mean ratings of 5 and above, non-emotional words had ratings between 3 and 4, while negative words had ratings below 3. The emotionality rating reflects the degree to which a word has strong emotional connotation. Words were selected as emotional if rated 5 or above, with emotionality ratings of words in the positive set matched with those in the negative set. This procedure was undertaken so as to match words in the negative and positive sets on the strength of their emotionality. The imagery rating indicates how readily a word evokes an imaged representation. Only words rated 4 or above were selected for inclusion in this study. Four separate one-way ANOVAs were computed on goodness, emotionality, imagery, and word frequency ratings (Kucera & Francis, 1967) for the combined positive and negative word sets. Goodness ratings for positive, negative and non-emotional stimuli were significantly different (F(3, 235) = 678.08; p < .001), with mean ratings for the positive and negative word sets 5.83 and 2.0, respectively. Mean goodness rating for the non-emotional words was 4.10 in the positive word set and 4.09 in the negative word set. The significant outcome for word emotionality (F(3, 235) = 457.26; p < .001) was further analysed with Tukey’s tests which indicated that the positive and negative sets were matched on emotionality as intended (with mean emotionality ratings 5.60 and 5.63, respectively) and they had significantly higher emotionality ratings than the nonemotional words in both the positive and negative word sets (with means of 2.7 and 2.54, respectively). There were no significant differences in imagery (F(3, 206) = 1.48; p = .22) or word frequency ratings (F(3, 230) = 0.86; p = .461) for stimuli in the positive and negative sets. Hence, words in the positive and negative sets were matched on the dimensions of emotionality, imagery and word frequency and differed on goodness ratings, as required. On this basis, we conclude that while ‘negative’ and ‘positive’ words were equated on level of emotionality, imageability and word frequency, they were clearly differentiated with respect to how intensely good or bad they are rated (i.e. their positive or negative connotation). All words were 3–8 letters in length and sets were matched on word length, as confirmed by a one-way ANOVA (F(3, 236) = 0.123; p = .947). It was also essential that the stimulus

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

sets designated for use in the primed and unprimed word conditions be matched on all word dimensions so that legitimate comparisons of hemispheric involvement in memory for primed stimuli could be made. Accordingly, independent samples t-tests computed on the primed and unprimed data sets for the positive and negative word sets indicated there were no significant differences on any word dimension in these conditions (all p > .05). Hence, the stimulus sets used in the primed and unprimed word conditions matched on the dimensions of goodness, emotionality, imagery, and word frequency. The materials for the perceptual encoding task consisted of typed booklets (one version for the positive word set and another for the negative set). Each booklet included an instruction sheet and a list of 60 words derived from Set 1 of each word valence group, with a blank box beside each word where participants recorded their response. Words appeared in a different random order in each printed booklet, to prevent order effects. As perceptual processing is significantly reduced by study-test modality shifts (Roediger & Srinivas, 1993), the size, type and letter case of the font closely matched the font characteristics of stimuli used in the lexical decision task. 2.1.5. Procedure Positive and negative words were presented in two different blocks in the test session, which lasted 45–50 min. Each block was preceded by the relevant encoding task. To operationalize the implicit nature of the memory component of the study (Roediger & McDermott, 1993), instructions for the perceptual encoding task referred to a language exercise, with no allusion to the priming component. First, participants completed the perceptual encoding task, by writing (in the box provided) the number of long straight lined strokes in each word. Next, they received instructions for the lexical decision task which emphasized the importance of minimizing errors and maintaining fixation on the central cross during stimulus presentation. No allusion was made to words from the encoding task reappearing in the lexical decision task. The instructions were followed by a practice block for the lexical decision task, with 30 words and 30 non-words, none of which were in the experimental set. No formal distracter tasks were used between the perceptual encoding and lexical decision tasks. In view of the length of the session, the number of task components involved, and inserting instructions and practice with different stimuli between perceptual encoding and lexical decision tasks, it was not considered necessary to also include a formal distracter task. Upon completion of the lexical decision task for the first block of stimuli (either the positive or negative block), the encoding task for the second block was completed, followed by the associated lexical decision task. In the lexical decision task, positive/negative and nonemotional stimuli that had been presented in the encoding task, and hence were perceptually primed, were presented with an equal proportion of unprimed stimuli. Each

1533

experimental trial began with the appearance of a central fixation which remained on screen for 500 ms. A word or non-word was then randomly projected to the left or right of the fixation, for 160 ms. The entire screen then remained blank for 1500 ms, during which the participant responded. Using a GO-NOGO response procedure, participants were required to simultaneously depress both index fingers on touch sensitive response keys when the target was a word, and to withhold their response for non-words. An “ERROR” message appeared directly above the central fixation for 300 ms if an incorrect response was made. The inter-trial interval was 500 ms. Upon completion of each block of 70 trials, participants were permitted to rest while feedback on their accuracy and mean RT for the preceding block was automatically displayed on the VDU. Participants completed a total of 210 trials for both the positive and negative word sets (i.e. a total of 420 trials) per session. 2.2. Results 2.2.1. Perceptual RT analyses First, it was necessary to ascertain whether order of presentation of the positive and negative word blocks interacted with any other variable, as this would indicate that order should be included as a variable in subsequent statistical analyses. Hence, statistical analyses were initially computed on mean correct RT for the perceptual data using a four-way split plot ANOVA, with order of presentation as the between-subjects variable and priming condition, word valence, and visual field as within-subjects variables. There was no main effect for order of block presentation (F(1, 27) = 0.53; p = .473). Only one interaction involving block order approached significance, and this was with valence (F(2, 54) = 3.07; p = .06). For participants who completed the positive word block first, there was little difference in RTs for words of differing valence (595 ms for positive words, 596 ms for negative words and 610 ms for non-emotional words). However, for participants who completed the negative block first, there was a relative advantage for positive words, which were responded to more quickly than negative and non-emotional words (at 555, 584 and 590 ms respectively). Hence, there was a tendency for one group of participants to respond more quickly to positive words, which were presented after the negative block of stimuli, relative to those who saw the negative block first. As this finding was not of major consequence in terms of the aims of the present study, and given that there was no main effect or other interactions involving order of presentation, subsequent statistical analyses for the perceptual data were collapsed across order. The perceptual data were then analysed to determine whether there were hemispheric differences in implicit memory for perceptually primed stimuli. A three-way repeated measures ANOVA (2 × 2 × 3) was computed with priming condition, visual field and word valence as variables. All main effects were significant. The main effect for priming condition (F(1, 28) = 22.84; p < .001) reflected the response

1534

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

facilitation of 21 ms for primed relative to unprimed words. The main effect for valence (F(2, 56) = 9.61; p = .001 with Huynh–Feldt correction), reflected faster responses to positive words (573 ms), relative to negative (590 ms) and nonemotional words (600 ms). The main effect for visual field (F(1, 28) = 53.83; p < .001) was due to a 34 ms facilitation for RVF relative to LVF presentations. There was only one significant interaction, this being between priming condition and visual field (F(1, 28) = 13.60; p = .001). Rather than focusing on this interaction, we incorporated valence along with priming condition and visual field into the next stage of analysis, as this allows assessment of our primary research question — whether implicit memory for perceptually primed words differs in each hemisphere as a function of word valence. To this end, separate two-way ANOVAs were computed for each visual field condition, with priming condition and valence as independent variables. In the RVF, while there was no interaction between prime condition and valence (F(2, 56) = 1.52; p = .228), both main effects were significant (as seen in Fig. 1). The main effect for prime condition (F(1, 28) = 30.124; p < .001) reflected the 35 ms RT advantage for primed relative to unprimed words. The main effect for valence (F(2, 56) = 4.404; p = .022 with Huynh–Feldt correction) reflected faster RTs to positive words (at 556 ms) relative to RTs for negative and nonemotional words which were 574 and 582 ms, respectively. Pairwise comparisons of these marginal means with least significant differences adjustment indicated that responses to positive words were significantly faster than responses to non-emotional words (p = .002). However, RT differences were not significant for positive and negative words, or negative and non-emotional words. Planned comparisons using paired samples t-tests were then computed for RTs to primed relative to unprimed words for each valence condition, to examine implicit memory in the left hemisphere for perceptually primed stimuli. Response latencies to primed stimuli were significantly faster than RTs to unprimed stimuli for positive (t(28) = 3.958; p < .001), negative (t(28) = 3.301;

p = .003) and non-emotional words (t(28) = 2.171; p = .039). This is suggestive of implicit memory in the left hemisphere for stimuli that had previously appeared in the encoding task, and that this implicit memory is independent of valence type. In the LVF, there was no interaction between prime condition and valence (F(2, 56) = .281; p = .756) and no main effect for prime condition (F(1, 28) = 1.828; p = .187), with RTs to primed words just 7 ms faster than RTs to unprimed words (at 601 and 608 ms). However, there was a main effect for valence (F(2, 56) = 5.841; p = .005) with RTs for positive, negative and non-emotional words being 591, 606 and 617 ms, respectively. Pairwise comparisons of the marginal means using least significant differences adjustment indicate that responses to positive words were significantly faster than RTs to non-emotional words (p = .001), while there was no significant difference in RTs for positive and negative words or for negative and non-emotional words (p > .05). Planned comparisons using paired samples t-tests failed to return any significant differences between RTs for primed relative to unprimed words for any valence condition (positive words: t(28) = 0.254; p = .801; negative words: t(28) = 1.101; p = .28; non-emotional words: t(28) = 0.376; p = .71). Evidently, there was no RT advantage for words that appeared in the encoding task and were subsequently presented in the LVF (see Fig. 1), although responses to positive words were generally faster than responses to negative or non-emotional words. 2.2.2. Perceptual error analyses Errors for the perceptual data were analysed with a threeway repeated measures ANOVA (2 × 2 × 3) with priming condition, visual field and valence as variables. The outcomes were consistent with the RT outcomes (refer to Table 1). Like the RT data, all main effects were significant. The main effect for priming condition (F(1, 28) = 57.92; p = .001) reflected less errors in responses to primed relative to unprimed words. The main effect for valence (F(2, 56) = 6.72; p = .002) reflected lower error rates for positive words, relative to negative and non-emotional words, while the main effect for

Fig. 1. Response latencies for perceptually primed and unprimed positive, negative and non-emotional words presented to the left and right visual fields.

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545 Table 1 Means (M) and standard deviations (S.D.) of error rates for perceptual data Word type

Word valence

Visual field of presentation LVF

RVF

M

S.D.

M

S.D.

Primed

Positive Negative Non-emotional

2.14 2.28 2.60

2.07 1.77 1.87

0.86 0.86 1.45

0.92 0.79 1.01

Unprimed

Positive Negative Non-emotional

2.93 3.97 3.55

2.02 2.61 2.00

1.14 2.07 2.14

1.09 1.65 1.39

visual field (F(1, 28) = 43.12; p < .001) indicated that errors were lower for RVF than LVF presentations. Hence, consistent with previous findings of left hemisphere superiority for lexical decision (Collins, 1999; Collins and Coney, 1993), responses were both faster and more accurate for RVF presentations. The only differences from the RT data were the two-way interaction between prime condition and valence (F(2, 56) = 4.936; p = .022) which reflected somewhat higher errors for unprimed negative words relative to positive and non-emotional words, and the absence of an interaction between prime condition and visual field. Even so, the error data followed a similar, albeit less pronounced pattern of response to the RT data. There were no speed/accuracy trade-offs. 2.3. Discussion We found an overall RVF advantage for lexical decisions, which is consistent with the established left hemisphere superiority on lexical decision tasks (Collins, 2002; Collins & Coney, 1990; Collins & Frew, 2001), as well as a general response facilitation in both visual fields for positive words relative to non-emotional words. However, word valence did not interact with visual field of presentation, and there was no interaction between prime condition and valence in either visual field. These findings mirror those of Nagae and Moscovitch (2002) who found an overall RVF advantage for their perceptual identification task, and no hemispheric differences in identifying words that varied in emotionality, and are consistent with their view that hemispheric differences in emotional processing are unlikely to emerge in tasks primarily tapping perceptual processes, and that such tasks are unhelpful in assessing the relative merits of the right hemisphere and valence models of emotional processing. Importantly, there was a RVF advantage in response latencies for words that had previously appeared in the perceptual encoding task, relative to new/unprimed words, but no difference in speed of response to primed and unprimed words presented to the LVF. This suggests that the previous encoding task engendered implicit memory of these words in the left hemisphere (which was not influenced by the valence of the words), while there was no evidence of implicit memory of perceptually encoded words in the right hemisphere. This may be interpreted within the framework

1535

developed by Marsolek (1995, 1999), Marsolek and Hudson (1999), Marsolek, Kosslyn, and Squire (1992) and Marsolek, Nicholas, and Andresen (2002) which outlines how visual forms, such as words, are perceived and stored in memory. Their research indicates that an abstract visual form system, which underlies recognition of abstract categories of shape (Marsolek & Hudson, 1999) and so can be used to identify particular words independently of their font or letter case (Marsolek, 1995) operates more effectively in the left hemisphere (Marsolek, 1995; Marsolek & Hudson, 1999). This system can be distinguished from a specific visual form subsystem, which preserves visual details when processing inputs (e.g. font) and operates more effectively in the right hemisphere (Marsolek, 1995; Marsolek & Hudson, 1999). Support for this model has been garnered from research measuring memory for visual forms presented in a divided visual field format (e.g. Marsolek, 1995; Marsolek et al., 2002) as well as studies using positron emission tomography which demonstrate that a region in the medial extrastriate cortex of the left hemisphere is activated when processing the visual structure of words (e.g. Peterson, Fox, Posner, Mintum, & Raichle, 1988). Interestingly, this region does not appear to be involved in processing the semantic aspects of words. Conceivably, the RVF implicit memory for stimuli primed according to their perceptual features in the current experiment reflects left hemisphere processing of the abstract visual representation of the word forms encountered in the encoding task which entailed judging the number of straight line strokes in these words. The absence of any impact of valence upon implicit memory in either hemisphere suggests that this implicit representation was not based upon the semantic features of these words.

3. Experiment 2 The second experiment primarily aimed to assess the contribution of each hemisphere to implicit memory for emotional words that have been conceptually encoded. Various factors were taken into consideration in designing appropriate encoding and implicit memory tasks for this purpose. First, the TAP approach highlights the importance of disentangling encoding and memory task requirements. Hence, in this experiment it was important to design an encoding task which elicited encoding of the semantic meaning of words, and then to pair it with a recall task which draws upon conceptual processes whilst minimizing the contribution of perceptual processes. Second, there is evidence that depth of processing manipulations, including generation of stimuli from associative cues, promotes conceptual processing of the semantic meaning of study items (Roediger & McDermott, 1993). Hence, we developed an encoding task which promoted the encoding of the semantic meaning of words. This conceptual encoding task required participants to generate an incomplete word from the cues provided in a sentence describing that word (e.g. another

1536

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

name for a house or place of residence: D··L···G). Then, based upon the principle that generating the semantic meaning of a given word automatically activates words which are associated in meaning (Collins & Coney, 1993, 1998; Neely, 1991), we reasoned that an implicit memory task reflecting the activation of words which are conceptually associated with the words previously generated in the encoding task would suit the requirements of this experiment. To this end, we developed an implicit memory task involving lexical decision, where half of the target words were conceptual associates of the words generated in the encoding task. Hence, all incomplete words in the encoding task (in the above example, DWELLING) were known associates of target words subsequently presented in the lexical decision task (e.g. ABODE). Therefore, the relationship between the encoding and retrieval of these words was based upon associative conceptual meanings. To preclude perceptually based recall, care was taken to ensure that the associates were not orthographically or phonologically similar to the incomplete words generated during encoding. Furthermore, the processing and response requirements of the encoding and recall tasks used in this experiment were thereby matched with those in Experiment 1, to allow direct comparisons of implicit memory following perceptual and conceptual encoding. Consistent with Nagae and Moscovitch’s (2002) argument that valence effects in each hemisphere are more readily detected when semantic meaning is analysed, we anticipated that conceptual encoding would differentially influence implicit memory for emotionally valenced words subsequently appearing in each visual field. In relation to the valence and RH models of emotional processing, support for the latter would be found in this experiment if RTs to emotional words presented in the LVF are faster than RTs to non-emotional words presented to this visual field. Support for the valence model would be indicated by a response advantage for positive words relative to negative and non-emotional words presented in the RVF, along with a response advantage for negative words relative to positive and non-emotional words presented to the LVF. 3.1. Method 3.1.1. Participants The participants from Experiment 1 returned for this experiment. As priming effects can carry-over when conceptual tasks are presented prior to perceptual tasks, whereas such effects are minimized when perceptual tasks precede conceptual tests (Graf & Mandler, 1984), all participants completed the conceptual experiment 7–10 days after the perceptual experiment. 3.1.2. Materials To optimize comparison of outcomes across the two experiments, and so be more confident that any RT differences could be attributed to the preceding encoding tasks rather than difference in stimulus words used, the same stimuli were

used in the lexical decision task in both experiments. Also, to reduce potential carry-over of priming effects from the encoding task in the first experiment, the encoding task for Experiment 2 was performed on a different subset of stimuli.2 The conceptual encoding materials consisted of two booklets: one each for the positive and negative word conditions. These contained an instruction sheet and 60 sentences, each of which provided letter cues for the generation of an incomplete word (e.g. a place where criminals are legally confined or incarcerated: J · · ·). All incomplete words were known word associates of target stimuli which were subsequently used in the primed condition of the lexical decision task (e.g. Jail-PRISON, where ‘prison’ is used in the lexical decision task). These word pairs were selected from word association norms (John, 1988; Shapiro & Palermo, 1968; Thomson, Meredith, & Browning, 1976). Sentences were constructed using definitions from the Macquarie dictionary (Delbridge, 1987) and Roget’s Thesaurus (1972) and arranged in a different random order in each booklet. Aside from the materials for the conceptual encoding task, this experiment was identical in all other respects to Experiment 1. 3.2. Results 3.2.1. Conceptual RT analyses Statistical analyses were computed on mean correct RTs for the lexical decision task. To determine whether order of presentation of the positive and negative word blocks interacted with any other variable, the conceptual data were initially analysed using a (2 × 2 × 2 × 3) four-way split plot ANOVA with block order as the between-subjects variable and priming condition, visual field and word valence as within-subjects variables. The main effect for block order approached significance (F(1, 27) = 3.9; p = .059), with RTs 55 ms faster for participants who completed the negative block first relative to those who completed the positive block first. There was an interaction between valence and block order (F(2, 54) = 6.45; p = .003): for participants who saw the positive block first, responses to positive words were faster than responses to negative and non-emotional words (at 578, 620 and 617 ms, respectively). However, for those who completed the negative block first, response latency was faster overall, and was equivalent for positive and negative words (at 544 ms) and 19 ms faster than latencies to non-emotional words. This is in keeping with research on the mobilization–minimization hypothesis that valence-based asymmetries in cognitive activity are due to the initial stronger physiological, emotional, and cognitive responses to negative stimuli relative to positive and non-emotional stimuli (Taylor, 1991). The general response advantage for both positive and negative words when the negative block was presented first differs from the findings of Experiment 2 These stimulus materials are also available at the website. http://www.psychology.murdoch.edu.au/publications/collins stimlistcandc. html.

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

1, where there was a (non-significant) tendency towards a RT advantage for positive words when they were presented after the negative block. As the same stimuli were presented in the lexical decision task in both experiments, it is likely that this discrepancy reflects differences stemming from the preceding encoding tasks. It appears that the conceptual encoding task facilitated encoding of valence related characteristics of these stimuli (as outlined below), unlike the perceptual encoding task where stimulus valence did not interact with visual field or priming condition. Consistent with the mobilization–minimization hypothesis (Taylor, 1991), after conceptual encoding which was designed to encourage semantic processing, there was a general response facilitation to all stimuli for participants who completed the negative block first, although this facilitation was greater for emotional words. While interesting, this outcome is not directly relevant to the experimental questions addressed in the current study. Moreover, interpretation of the conceptual data is not compromised by the effects of block order, as this variable did not interact with either prime condition or visual field. Thus, all subsequent analyses collapsed across block order as a variable. The conceptual data were then analysed to determine whether the conceptual encoding task influenced implicit memory in each hemisphere for conceptually primed stimuli. A three-way repeated measures ANOVA (2 × 2 × 3) was computed with priming condition, visual field and word valence as variables. There were two main effects. The main effect for valence (F(2, 56) = 10.15; p = .002; with Huynh–Feldt correction), reflected the faster RTs to positive words (559 ms) relative to negative (578 ms) and nonemotional words (588 ms). The main effect for visual field (F(1, 28) = 37.48; p < .001) reflected the 34 ms advantage for words presented in the RVF, and again replicates the left hemisphere advantage for lexical decisions (e.g. Collins, 1999, 2002). Although there was no three-way interaction, all of the two-way interactions were significant: priming condition and valence (F(2, 56) = 6.55; p = .003); priming condition and visual field (F(1, 28) = 4.85; p = .036); valence and visual field (F(2, 56) = 3.62; p = .033). Subsequent analyses focused on identifying the source of each of these interactions. The interaction between priming condition and visual field was explored with related samples t-tests comparing RTs for primed and unprimed words in each visual field. There was no significant difference in RTs for primed and unprimed words presented in the LVF (t(28) = 1.007; p > .05). However, in the RVF there was a significant advantage (of 9 ms) for primed words relative to unprimed words (t(28) = 2.238; p = .033). This is suggestive of implicit memory in the left hemisphere for conceptually primed information. Next, to isolate the source of the interaction between valence and visual field, separate one-way ANOVAs were computed for each visual field with valence as the independent variable. This analysis revealed a significant effect for valence in both visual fields (RVF: F(2, 56) = 8.04; p = .002; LVF: F(2, 56) = 8.29; p = .002, both with Huynh–Feldt

1537

correction). The source of these significant effects was traced with paired samples t-tests comparing mean RTs for positive, negative, and non-emotional words in each visual field. In the LVF, responses to positive words were significantly faster than responses to both negative (t(28) = 3.01; p = .005) and non-emotional words (t(28) = 4.99; p < .001), but there was no significant difference in latencies for negative and non-emotional words (t(28) = 0.114; p > .05). The RT advantage for positive words relative to negative and nonemotional words presented in the LVF suggests that the right hemisphere’s response to words varied as a function of emotional valence, and may be interpreted as support for the RH model. For RVF presentations, responses to positive words were significantly faster than responses to non-emotional words (t(28) = 4.71; p < .001), just as responses to negative words were faster than to non-emotional words (t(28) = 3.13; p = .004). However, there was no significant difference in RTs for positive and negative words (t(28) = 1.027; p = .313). Evidently, emotional words projected to the left hemisphere were processed more quickly than non-emotional words. Examination of the significant interaction between valence and visual field provides limited insight into the nature of hemispheric contributions to implicit memory for conceptually primed stimuli, as implicit memory for these stimuli remains obscured if responses for primed and unprimed words remain undifferentiated in the statistical analysis. Since the primary aim of this experiment was to examine implicit memory in each hemisphere for both perceptually and conceptually primed relative to new/unprimed stimuli, at the next stage of the analysis we incorporated the data from both experiments, in order to differentiate between predictions for the right hemisphere and valence models. 3.2.2. Combined RT analyses of Experiments 1 and 2 A four-way repeated measures ANOVA (2 × 2 × 2 × 3) was computed with experiment, priming condition, visual field and word valence as variables. There were main effects for prime condition (F(1, 28) = 17.636; p < .001), visual field (F(1, 28) = 64.701; p < .001) and valence (F(2, 56) = 22.904; p < .001 with Huynh–Feldt correction). Hence, while there was no overall difference in RTs for Experiments 1 and 2, as would be expected given that the same targets were presented in the lexical decision task for both experiments (F(1, 28) = 2.293; p = .141), responses were faster overall for primed relative to unprimed words (575 cf. 587 ms), just as responses were significantly faster (by 34 ms) for RVF presentations. Also, responses were faster for positive words (566 ms) relative to negative (584 ms) and non-emotional words (594 ms). However, these main effects were mediated by significant interactions between experiment and prime condition (F(1, 28) = 14.022; p = .001), prime condition and visual field (F(1, 28) = 20.681; p < .001), and experiment, prime condition and valence (F(2, 56) = 5.866; p = .005). Although there was no four-way interaction involving visual field and the latter three variables (F(2, 56) = 1.642; p = .203), we incorporated visual

1538

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

field in subsequent analyses, as dictated by the experimental aims of the current study which were to examine implicit memory in each hemisphere for emotional and non-emotional words after perceptual and conceptual encoding. To this end, separate three-way ANOVAs were computed for each visual field, with experiment, prime condition and valence as variables. 3.2.3. RVF combined analyses In the RVF, the main effects for prime condition (F(1, 28) = 37.809; p < .001) and valence (F(2, 56) = 11.93; p < .001 with Huynh–Feldt correction) were significant. There was a 22 ms RT advantage for primed relative to unprimed words, while RTs to positive, negative and non-emotional words were 550, 564 and 578 ms, respectively. Pairwise comparisons of valence computed with least significant differences adjustment indicated that responses to emotional words were significantly faster than responses to non-emotional words (positive cf. non-emotional words: p < .001; and negative cf. non-emotional words: p = .008) but the difference in RT for positive and negative words was not significant (p = .062). There were two significant interactions, these being between experiment and prime condition (F(1, 28) = 11.506; p = .002) and experiment, prime condition and valence (F(2, 56) = 4.997; p = .01). Identification of the source of the threeway interaction entailed separate analysis of prime condition and valence for each experiment. As this analysis has already been computed for the perceptual experiment, a twoway ANOVA was then computed for RVF presentations in the conceptual experiment, with prime condition and valence as independent variables (as depicted in Fig. 2). In the conceptual experiment, the main effect for prime (F(1, 28) = 5.01; p = .033) reflected the 11 ms RT advantage for primed relative to unprimed words presented in the RVF. There was also a main effect for valence (F(2, 56) = 8.04; p = .002; with Huynh–Feldt correction) with RTs fastest for positive words (545 ms), slowest for non-emotional words (574 ms) and intermediate for negative words (554 ms). The interaction between prime condition and valence (F(2,

56) = 4.17; p = .02) was further analysed with separate oneway ANOVAs for primed and unprimed RVF presentations, with valence as the independent variable. The ANOVA for unprimed words was significant (F(2, 56) = 9.15; p < .001), with RTs to non-emotional words (at 588 ms) considerably slower than RTs for positive and negative words (548 and 550 ms, respectively). These data were further analysed using paired samples t-tests to compare responses for each valence type. Responses to unprimed positive and negative words were faster than responses to non-emotional words (t(28) = 4.096; p < .001; and t(28) = 4.243; p < .001, respectively), while there was no significant difference in RTs for unprimed positive and negative words (t(28) = 0.179; p > .05). In contrast, the ANOVA for primed words was not significant (F(2, 56) = 2.65; p > .05), indicating that the left hemisphere responded similarly to primed words of differing valence. Lastly, planned comparisons using paired samples t-tests comparing differences in RTs for primed relative to unprimed words for each word valence were computed to examine implicit memory in the left hemisphere for conceptually primed stimuli. These comparisons were used to assess whether conceptual encoding differentially impacted upon the implicit memory of emotional words presented in the RVF, and allowed us to evaluate the veracity of the right hemisphere and valence models. These t-tests indicated that the 28 ms response advantage for primed relative to unprimed nonemotional words was significant (t(28) = 4.07; p < .001). In contrast, there were no significant differences in latencies for primed and unprimed positive words (t(28) = 0.710; p > .05), or for primed and unprimed negative words presented in the RVF (t(28) = 0.968; p > .05). As seen in Fig. 2, this suggests that conceptual encoding engendered implicit memory in the left hemisphere for non-emotional words, but not for positive or negative words. 3.2.4. LVF combined analyses In the LVF, the three-way ANOVA computed with experiment, prime condition and valence as variables, returned only one significant effect — a main effect for valence

Fig. 2. Response latencies for conceptually primed and unprimed positive, negative and non-emotional words presented to the left and right visual fields.

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

(F(2, 56) = 15.765; p < .001). Response latencies to positive words (at 582 ms) were significantly faster than responses to both negative (at 604 ms, p = .001) and non-emotional words (at 609 ms, p < .001). As there was a trend towards an interaction between experiment, prime condition and valence (F(2, 56) = 2.603; p = .083) and as this study aimed to ascertain whether the right hemisphere is involved in implicit memory for words of differing valence, additional analyses (paralleling those completed for the RVF above) entailed a two-way ANOVA computed for the conceptual experiment, with prime and valence as the independent variables (the equivalent ANOVA was previously computed for the perceptual experiment). There was only one significant effect — a main effect for valence (F(2, 56) = 8.29; p = .002, with Huynh–Feldt correction). RTs were faster for positive words (573 ms) relative to negative and non-emotional words (602 and 601 ms, respectively). There was a trend towards an interaction between prime condition and valence in the conceptual experiment (F(2, 56) = 2.85; p = .079; with Huynh–Feldt correction) as depicted in Fig. 2. To assess this trend, and thereby evaluate the right hemisphere’s involvement in the processing and implicit memory of conceptually primed emotional words, separate one-way ANOVAs were computed for conceptually primed and unprimed words presented in the LVF, with valence as the independent variable. There was no significant effect for unprimed words (F(2, 56) = 2.048; p > .05), indicating that RT to new words did not differ as a function of emotional valence. However, there was a significant effect for primed words (F(2, 56) = 12.07; p < .001), with RTs to positive words (569 ms) considerably faster than latencies for non-emotional words (601 ms) and negative words (614 ms). This was further analysed using paired samples t-tests, which revealed that responses to positive primed words were significantly faster than responses to both negative (t(28) = 3.946; p < .001) and non-emotional primed words (t(28) = 4.641; p < .001). However, the difference in RTs for primed negative and nonemotional words was not significant (t(28) = 1.384; p > .05). Then, consistent with the RVF data analysis, planned comparisons using paired samples t-tests were computed on RTs for primed and unprimed words presented in the LVF to examine implicit memory in the right hemisphere for conceptually primed stimuli. The 23 ms response inhibition for primed relative to unprimed negative words was significant (t(28) = 2.08; p = .047). However, RTs for primed and unprimed words did not differ significantly for either positive (t(28) = 1.029; p > .05) or non-emotional words (t(28) = 0.104; p > .05). It seems that the conceptual encoding task engendered implicit memory in the right hemisphere for negative words, but not for positive or non-emotional words. In summary, the combined RT analyses for Experiments 1 and 2 are indicative of a hemispheric dissociation in patterns of implicit memory for words primed according to their perceptual and semantic features. Although in both experiments there was a general RT advantage for positive words relative to negative and non-emotional words, there were also clear

1539

visual field differences in response latencies for words of differing valence that was moderated by priming condition. In Experiment 1, there was a RVF advantage for words that had previously appeared in the perceptual encoding task, relative to new/unprimed words, and this effect was independent of word valence. However, relative to unprimed words, there was no RT advantage for words appearing in the perceptual encoding task and subsequently presented in the LVF. Hence, after the perceptual encoding task, implicit memory of previously encountered words was evident in the left hemisphere alone. In Experiment 2, for RVF presentations, responses to unprimed emotional words were faster than responses to unprimed non-emotional words. Conversely, valence did not differentially impact upon responses to unprimed words presented in the LVF. When responses for primed and unprimed words were compared (as a measure of implicit memory), there were clear visual field differences that were moderated by word valence. For RVF presentations, responses to nonemotional associates of the words that had been presented in the conceptual encoding task were significantly faster than responses to new/unprimed non-emotional words. In contrast, there was no response advantage for emotional associates of the words presented in the encoding phase relative to new emotional words. For LVF presentations, responses to negative words involved in the encoding phase were significantly slower than responses to new/unprimed negative words, but there was no difference in latencies for primed relative to unprimed positive or non-emotional words. These outcomes suggest that conceptual priming engendered implicit memory for non-emotional words in the left hemisphere, and implicit memory for negative words in the right hemisphere. 3.2.5. Conceptual error analyses Error rates for the conceptual data were analysed by computing a three-way repeated measures ANOVA (2 × 2 × 3) with priming condition, visual field and word valence as variables. The outcomes were consistent with the corresponding RT analysis. There were main effects for valence (F(2, 56) = 9.47; p < .001) and visual field (F(1, 28) = 23.66; p < .001). As seen in Table 2, errors were lowest for positive words, with similar error rates for negative and non-emotional words. Errors were also lower for stimuli presented to the RVF. Hence, responses were both faster Table 2 Means (M) and standard deviations (S.D.) of error rates for conceptual data Word type

Word valence

Visual field of presentation LVF

RVF

M

S.D.

M

S.D.

Primed

Positive Negative Non-emotional

1.86 3.42 2.21

1.60 2.15 1.62

1.24 1.55 1.28

1.15 1.33 1.02

Unprimed

Positive Negative Non-emotional

1.97 2.28 2.88

1.57 1.94 1.99

1.00 1.00 1.30

1.00 1.04 1.20

1540

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

and more accurate for stimuli presented in the RVF. There were interactions between prime condition and valence (F(2, 56) = 0.375; p < .001), as well as valence and visual field (F(2, 56) = 4.83; p = .012). The error data for these interactions mirrored the pattern of response in the RT data. There were no speed/accuracy trade-offs in the data.

4. General discussion Interpretation of the current results hinges on whether the encoding tasks elicited perceptual and conceptual processing as anticipated. Comparison of the results for Experiments 1 and 2, confirms that the encoding tasks employed for each experiment differentially affected the manner in which participants processed emotionally valenced stimuli. In Experiment 1, the general response facilitation for perceptually primed words relative to unprimed words, in conjunction with the absence of any hemispheric differences in responses to primed words of differing valence, suggests that engagement in the perceptual encoding task facilitated retrieval of the surface/perceptual features of these words and not their emotional meanings. In contrast, in Experiment 2, the hemispheric differences in response to words which had been primed according to their semantic meaning relative to unprimed words, in combination with the hemispheric differences in responses to words of differing valence, is consistent with the conclusion that engagement in the conceptual encoding task differentially influenced the manner in which each hemisphere was involved in the processing and memory of emotionally valenced stimuli. Moreover, the hemispheric dissociation in patterns of implicit memory for words primed according to either their perceptual or semantic features, as reflected in the interactions between experiment, prime condition and valence in the combined analyses, bolsters the conclusion that the two encoding tasks were effective in eliciting different modes of processing, even though a lexical decision task using the same stimuli was used as a measure of implicit memory in both cases. Taken together, these results unambiguously support the conclusion that the encoding tasks elicited perceptual and conceptual processing as intended. This functional dissociation between performance associated with the perceptual and conceptual implicit memory tasks is consistent with the TAP account of memory which assumes that this dissociation is invoked by the different modes of processing employed during encoding and retrieval (Roediger & McDermott, 1993). In Experiment 1, the 35 ms advantage on the lexical decision task for perceptually primed relative to unprimed words appearing in the RVF, suggests that the encoding task engendered an implicit memory in the left hemisphere for the perceptual features of words. We found no evidence of implicit memory in the right hemisphere for the perceptual features of words, as there was no difference in response latencies for primed and unprimed words presented in the LVF. These findings are consistent with recent evidence that an abstract

visual word form subsystem, which processes and stores abstract representations about the visual structure of word forms but not the semantic or conceptual information associated with them (Marsolek & Hudson, 1999; Schacter, 1992), operates more effectively in the left hemisphere (Marsolek, 1995, 1999; Marsolek et al., 1992). Hence, consistent with the view that primed information is retrieved from memory as a result of heightened activation of item specific information (Graf & Mandler, 1984; MacLeod & Bassili, 1989) participation in the perceptual encoding task appears to have activated modality specific information stored in the abstract visual word form subsystem in the left hemisphere, and subsequently rendered it more accessible to memory during RVF presentations in the lexical decision task. Importantly, although there was a general advantage for positive words in both visual fields, implicit memory of words primed according to their perceptual features was not influenced by word valence. Hence, these results are consistent with Nagae and Moscovitch’s (2002) stance that hemispheric differences in emotional processing are unlikely to emerge in tasks primarily tapping perceptual processes, and accounts for the absence of valence effects in tasks primarily tapping implicit memory for perceptual information (e.g. Coney & Fitzgerald, 2000; Eviatar & Zaidel, 1991; Watkins, Martin, & Stern, 2000). Evidently, perceptually based tasks are of limited value in evaluating the relative merits of the right hemisphere and valence models of emotional processing. In contrast, a number of interesting hemispheric differences were obtained after conceptual encoding, and these shed light on the veracity of the right hemisphere and valence models of emotional processing. When interpreting these data, it is important to be mindful that hemispheric processing of emotional words is primarily reflected in lexical decision responses for the unprimed condition, while co-consideration of responses for the primed and unprimed conditions provides insight to implicit memory for emotional words. In relation to the former, emotional words were responded to more quickly than non-emotional words in the RVF, but there was no difference in response latency for positive and negative words (as depicted in Fig. 2). This suggests that the left hemisphere processed emotional words more quickly than non-emotional words, which is inconsistent with the valence model’s prediction of a RVF advantage for positive but not negative words. Nevertheless, the general processing advantage for emotional stimuli presented to the left hemisphere is consistent with evidence that selective attention is normally directed towards emotional stimuli due to their high behavioural significance (Compton, Heller, Banich, Palmieri, & Miller, 2000). For LVF presentations, responses to words of differing valence were moderated by previous conceptual encoding. Latencies for unprimed words did not differ as a function of emotionality. However, for words primed according to their semantic meaning, responses to positive words were considerably faster than responses to both negative and non-emotional words (by 32 and 45 ms, respectively). This indicates that there was a processing advantage in the right hemisphere

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

for primed positive words relative to words of negative and neutral valence. However, there was no evidence of implicit memory for positive words in the right hemisphere, as there was no corresponding advantage for primed relative to unprimed positive words appearing in the LVF. It seems that by presenting all positive words in the same block, we effectively elicited a ‘positive bias’ which facilitated responses to positive words, but did not also elicit implicit memory of previously encoded associates. Given the general response advantage for positive words relative to negative and nonemotional words in both visual fields for both experiments, it is also possible that the characteristics of the positive words per se may have promoted faster responses than the negative and non-emotional stimuli, despite our careful attempts to match all stimuli on dimensions likely to differentially impact upon responses. This must be considered in light of considerable evidence that positive emotional stimuli are perceived more easily and are associated with higher levels of recall than negative stimuli (Ali & Cimino, 1997; Cacioppo et al., 1985; Riskind & Lane, 1987) which suggests that positive stimuli are recognized more easily and this would also be reflected in faster responses to positive words relative to negative and non-emotional words. Comparison of responses for primed and unprimed words presented in each visual field provides a measure of implicit memory for emotional words in each hemisphere. While we found evidence of implicit memory restricted to the left hemisphere for the perceptual features of words, the absence of word valence effects suggests that this implicit representation did not incorporate the emotional characteristics of these words. In contrast, after conceptual encoding of matched stimuli there is evidence of left hemisphere implicit memory for non-emotional words and right hemisphere implicit memory for negatively valenced words. Responses to LVF presentations of non-emotional and positive words that were primed according to their semantic meaning did not differ from responses to new (unprimed) non-emotional and positive words. However, responses to conceptually primed negative words were significantly slower than responses to new (unprimed) negative words. Hence, responses to negative words were inhibited after conceptual encoding of associates of those words. It is of interest here that responses to negative primed words presented in the RVF were also somewhat slower (albeit not significantly) than responses to unprimed negative words (by 9 ms). This inhibition can be accounted for within the framework of inhibitory models of processing which posit that specific cognitive mechanisms “selectively inhibit the internal representation of mental constructs” (Williams et al., 1997, p. 23). The operation of these inhibitory cognitive mechanisms is apparent in priming tasks as “response impairments” (Williams et al., 1997, p. 24) for related targets. This ‘negative priming effect’ is often observed in the inhibition of responses to emotionally threatening inputs (Williams et al., 1997). Such inputs include stimuli of negative emotional valence. Within the context of the current experiment, it appears that the mechanisms

1541

which selectively inhibit internal representations of negatively valenced stimuli are particularly active in the right hemisphere. This is consistent with several previous findings of interference effects in the right hemisphere for positive and negative stimuli (Compton et al., 2000; Hartikainen, Ogawa, & Knight, 2000; Richards et al., 1995) which have been interpreted as evidence of the right hemisphere’s preferential involvement in processing emotional stimuli. Hence, support for the RH model is garnered by either facilitation or inhibition of response to emotional stimuli presented in the LVF. In this vein, the inhibition of associates of previously encoded negatively valenced words in the current study is indicative of right hemisphere involvement in implicit memory of negatively valenced emotional stimuli, and more generally with evidence that the right hemisphere is involved in withdrawal responses (Davidson, 1992, 1998). The left hemisphere’s involvement in implicit memory for conceptually primed words in the current study was quite different to the pattern observed for the right hemisphere. Prior conceptual encoding did not facilitate responses to associated positive or negative words subsequently appearing in the RVF. Hence, although new (unprimed) emotional words were processed more quickly than non-emotional words in the RVF, previous conceptual encoding of emotional associates did not then enhance their implicit memory, even though positive and negative stimuli were presented in separate blocks to encourage a valence specific bias. This could be interpreted as an indication that the processing benefit gained from emotionality and priming is not additive. That is, responses may be facilitated either by the emotionality of a word, or by previous presentation of its associates, but once responses to complex stimuli reach maximal speed, there is no added benefit when a word is both primed and emotional.3 Comparison of the response speed in the current study (which varied within a range of 541–614 ms) with previous studies which have also used lexical decision with stimuli presented in a divided visual field format, permit us to evaluate whether RTs in the current study were of maximal speed, which thereby obviated any additional processing advantage achievable by combining priming and emotionality. In a study measuring speed of lexical decision responses to letter strings presented in the LVF and RVF, Coney and Collins (1994) obtained mean RTs between 396 and 506 ms, with mean RT for most conditions being 420–450 ms. Further, studies which have used lexical decision and the divided visual field format, but increased the complexity of stimulus presentation by incorporating priming stimuli, have regularly found faster responses (around 450–500 ms) than those in the current study, despite using the same experimental hardware and software (e.g. Collins, 1999; Collins & Coney, 1998). Burgess and Simpson (1988) also obtained mean RTs between 392 and 518 ms, although mean RTs for most of their conditions were under 450 ms. The consistently higher mean error rates in those studies relative to the current study confirm that their 3

We would like to thank an anonymous reviewer for this suggestion.

1542

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

task requirements were more difficult: mean errors in Collins and Coney’s study (1998) ranged between 8 and 24%, dependent upon visual field and word pair relationship, while mean errors in Burgess and Simpson’s study were 10–20%. Errors in the current study were considerably lower, with a range of 3.3–11.4% (refer to Table 2). As error rates reflect task difficulty, the lower error rates in the current study confirm that the task we used was considerably less complex than similar studies which have obtained consistently faster RTs. Hence, it is unlikely that RTs in the current study were of maximal possible speed, and that this then precluded additional facilitation for primed emotional words presented in the RVF. This conclusion is also consistent with current understanding of the mechanisms thought to underlie facilitation of responses to emotional and primed stimuli: the processing advantage for emotional stimuli is normally associated with mechanisms of selective attention (Compton et al., 2000) while the mechanisms for priming in implicit memory tasks is viewed as a reflection of activation of item specific information (MacLeod & Bassili, 1989) or activation of associates in memory (Bower, 1981). As these mechanisms contribute separately to facilitation of responses to stimuli that are both emotional and primed, it seems unlikely that maximal speed achieved by one of these mechanisms will obviate additional facilitation potentially contributed by the other mechanism. Even if this were possible, it is unlikely to apply in the current study, as comparable studies outlined above indicate that the responses obtained are unlikely to be of maximal potential speed. Given this, a reasonable interpretation of the absence of a response advantage for conceptually primed emotional words relative to unprimed emotional words presented to the RVF is that the left hemisphere is not involved in implicit memory for emotional words. It is of interest then, that the results of Experiment 2 indicate the left hemisphere is involved in implicit memory of non-emotional stimuli. This conclusion is based on the observation that responses to non-emotional words presented to the RVF were facilitated following conceptual encoding relative to response latencies for new nonemotional words. It appears that the left hemisphere is specifically involved in implicit memory for non-emotional words and in the initial recognition of words with emotional connotation. More generally, while certain aspects of the findings of Experiment 2 appear to support the valence model, the overall pattern of results is more consistent with the RH model. The response advantage for positive words presented to the RVF relative to the LVF could be viewed as sensitivity of the left hemisphere to positive emotion. However, this finding must be viewed within the context of the general advantage for words presented in the RVF in the current study, which is consistent with a left hemisphere advantage for lexical decision (Collins, 1999, 2002) and is likely to reflect the central role of the left hemisphere in word identification (Nagae & Moscovitch, 2002). As there was no difference in response latencies to negative and positive words presented in the RVF, both

of which were faster than responses to new non-emotional words, these findings are indicative of left hemisphere sensitivity to emotional connotation that is not specific to positive valence. Since this pattern was consistent across the primed and unprimed conditions, and there was no accompanying evidence of implicit memory for emotional words presented in the RVF, this sensitivity to emotional connotation appears to reflect an alerting function for stimuli of behavioural significance (Compton et al., 2000). The inhibition of responses to primed negative words and the absence of priming for positive words presented to the LVF in Experiment 2 are both consistent with the valence model’s assumption that the right hemisphere is sensitive to negative, but not positive emotion. However, other aspects of the current findings are inconsistent with this interpretation. Specifically, there was no indication that the right hemisphere was differentially sensitive to negative words relative to positive or non-emotional words, as there was no valence effect for unprimed words presented to the LVF. It is interesting in this context that new emotional words (both negative and positive) were responded to more quickly than new non-emotional words presented to the RVF. Both of these findings are inconsistent with the valence model’s prediction of differential sensitivity to positive emotion in the left hemisphere and negative emotion in the right hemisphere. Moreover, for primed words presented to the LVF, responses were considerably faster for positive words than non-emotional and negative words. This signifies a right hemisphere sensitivity to positive emotion, even though there was no associated implicit memory for positive words. In sum, the findings of Experiment 2 are consistent with the contention that the right hemisphere mediates the processing of both positive and negative emotion, and so support the RH model of emotion. This conclusion is based upon the general LVF facilitation in responses to positive words relative to words of negative and non-emotional valence, and by the 23 ms inhibition for LVF presentations of conceptually encoded negative words relative to new negative words. Evidently, the right hemisphere is involved in forming implicit memories of negatively valenced emotional material and is generally responsive to positive valence. In contrast, the left hemisphere is involved in implicit memory for non-emotional words and is generally responsive to emotional connotation, which appears to act as an alerting function to stimuli of behavioural significance. Importantly, comparison of the results of Experiments 1 and 2 confirm that hemispheric differences in emotional processing are more likely to manifest at stages of processing associated with analysis of the emotional meaning of stimuli: we found no evidence of hemispheric differences in responses to the emotionality of words when encoding focused on their surface/perceptual features, yet there were clear hemispheric differences when encoding focused on their semantic/conceptual meaning. Similarly, Nagae and Moscovitch (2002) found evidence consistent with the RH model when using a memory task which tapped explicit recall and the

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

conceptual processing of stimuli, but not with a task tapping perceptual identification of words. Conceivably, the right hemisphere mediates both explicit and implicit memories of emotional material. However, it is of interest that Ali and Cimino (1997, 1998) found support for the valence model in studies that also examined hemispheric lateralization of memory for emotional words. The discrepancy in these findings may be attributable to important methodological differences: while the current study employed a lexical decision task as the measure of implicit memory, Ali and Cimino used lexical decision as their encoding task. Moreover, we adopted a priming paradigm to measure implicit memory, with reaction time as the dependent variable, while Ali and Cimino adopted a standard memory paradigm where they measured free recall of words previously presented in a divided visual field format followed by delayed recognition of these words. They also used a stimulus set with far fewer items than the current study, which necessitated multiple exposures of stimuli within each visual field, and is likely to have impacted upon memory for these items. Moreover, we ensured that stimuli within and across the two experiments were matched on a variety of dimensions known to be relevant to emotional processing, while Ali and Cimino did not. We also presented positive and negative stimuli in separate blocks, which is likely to have maximized valence effects in a manner consistent with studies manipulating mood through repeated exposure to positively or negatively valenced statements (see Williams et al., 1997), while Ali and Cimino intermixed words of different valence. Future research focusing upon the impact of mood induction and the processing of word emotionality by each cerebral hemisphere would clarify the impact of participants’ emotional state upon the variables examined in these studies. In conclusion, the current study found a clear hemispheric dissociation in implicit memory for words encoded perceptually or conceptually. This dissociation substantiates the benefit of using a TAP approach to evaluate memory for emotional stimuli in each hemisphere. It also confirms the TAP model’s assumption that depth of processing manipulations will impact upon conceptual memory performance on both implicit and explicit tasks (Roediger et al., 1989; Vaidya et al., 1997). Although many previous studies have found this assumption difficult to substantiate when using implicit memory tasks which manipulate conceptual elaboration (MacLeod & Bassili, 1989), the present study found clear evidence that conceptual encoding impacted upon implicit memory, and this was demonstrated in a lexical decision task which favoured automatic as opposed to strategic processing. Little is known about the mechanisms underlying conceptual priming (Vaidya et al., 1997), although it has been hypothesised that the priming observed in implicit memory tests is primarily a reflection of activation of item specific logogen information (MacLeod & Bassili, 1989). While this view may account for performance on tests of perceptual implicit memory, it does not account for the current findings, as we presented different words during encoding and retrieval for the conceptual encoding condition, and the relationship

1543

between encoding and retrieval was based solely upon association of semantic meaning. It is more likely that the priming observed in the current study for conceptual associates of words presented during encoding, reflects activation of associated concepts within memory (see Bower, 1981) rather than activation of item specific logogen information. Conceptual encoding seems to have facilitated access to associates subsequently presented in the lexical decision task, with these associates retrieved via automatic processes. The most interesting aspect of the current findings was the hemispheric dissociation in implicit memory for emotional stimuli. We found clear evidence that perceptual and conceptual encoding differentially impacts upon the formation of implicit memory in each cerebral hemisphere for emotionally valenced words. In doing so, we have demonstrated the usefulness of adopting a TAP approach to investigate memory in each hemisphere, and concurrently provided support for Nagae and Moscovitch’s (2002) argument that hemispheric differences in emotional processing are more likely to manifest at stages of processing when analysis of emotional meaning takes place. More specifically, we found that the left hemisphere recognizes new emotional words more rapidly than non-emotional words. While similar outcomes in previous studies have been interpreted as support for the valence model (e.g. Ali & Cimino, 1998; Coney & Fitzgerald, 2000; Vaidya et al., 1997), the findings of the current study and those of Nagae and Moscovitch (2002) suggest that it is more likely that those findings reflect the left hemisphere advantage for word identification. Instead, the current study indicates that while the left hemisphere is involved in initially recognizing that words have an emotional connotation, it does not mediate implicit memories of these emotional connotations. In contrast, in tasks engendering conceptual processing, the right hemisphere is involved in forming implicit memories of negatively valenced words, and is generally responsive to positive emotional valence. Together, the findings of the current study provide support for the RH model of emotional processing, and indicate that while the left hemisphere mediates implicit memory of non-emotional linguistic stimuli, the right hemisphere mediates implicit memory of negatively valenced linguistic material and is generally responsive to positive emotion.

References Ali, N., & Cimino, C. R. (1997). Hemispheric lateralization of perception and memory for emotional verbal stimuli in normal individuals. Neuropsychology, 11, 114–125. Ali, N., & Cimino, C. R. (1998). Hemispheric lateralization of memory for verbal emotional stimuli: Accounting for both valence and arousal dimensions. Unpublished doctoral dissertation, University of South Florida, Tampa. Bock, M. (1986). The influence of emotional meaning on the recall of words processed for form or self-reference. Psychological Research, 48, 107–112. Borod, J. C. (1992). Interhemispheric and intrahemispheric control of emotion. Journal of Consulting and Clinical Psychology, 60, 339–348.

1544

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545

Borod, J. C., Andelman, F., Obler, L., Tweedy, J., & Welkowitz, J. (1992). Right hemisphere specialization for the identification of emotional words and sentences: Evidence from stroke patients. Neuropsychologia, 30, 827–844. Borod, J. C., Cicero, B., Obler, L., Welkowitz, J., Erhan, H., Santshi, C., et al. (1998). Right hemisphere emotional perception: Evidence across multiple channels. Neuropsychology, 12, 446–458. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129–148. Brody, N., Goodman, S., Halm, E., Krinzman, S., & Sebrechts, M. (1987). Lateralized affective priming of lateralized affectively valued target words. Neuropsychologia, 25, 935–946. Brown, W. P., & Ure, D. (1969). Five rated characteristics of 650 word association stimuli. British Journal of Psychology, 60, 233–249. Bryden, M. P. (1977). Measuring handedness questionnaire. Neuropsychologia, 15, 617–624. Burgess, C., & Simpson, G. (1988). Cerebral hemispheric mechanisms in the retrieval of ambiguous word meanings. Brain and Language, 33, 86–103. Cacioppo, J., Petty, E., & Morris, K. (1985). Semantic, evaluative, and self-referent processing: Memory, cognitive effort, and somatovisceral activity. Psychophysiology, 22, 371–384. Cicone, M., Wapner, W., & Gardner, H. (1980). Sensitivity to emotional expressions and situations in organic patients. Cortex, 16, 145–158. Collins, M. A. (1999). Differences in semantic category priming in the left and right cerebral hemispheres under automatic and controlled processing conditions. Neuropsychologia, 37, 1071–1085. Collins, M. A. (2002). Interhemispheric communication via direct connections for alternative meanings of ambiguous words. Brain and Language, 80, 77–96. Collins, M. A., & Coney, J. (1990). Semantic and phonemic priming in the cerebral hemispheres. Neuropsychologia, 28, 933–945. Collins, M. A., & Coney, J. (1993). Associative priming in the cerebral hemispheres as a function of SOA. Neuropsychologia, 31, 1397– 1409. Collins, M. A., & Coney, J. (1998). Interhemispheric communication is via direct connections. Brain and Language, 64, 28–52. Collins, M. A., & Frew, A. (2001). Right hemisphere contributions to the comprehension of low imagery words. Laterality, 6, 111–132. Compton, R., Heller, W., Banich, M., Palmieri, P., & Miller, G. (2000). Responding to threat: Hemispheric asymmetries and interhemispheric division of input. Neuropsychology, 14, 254–264. Coney, J., & Collins, M. A. (1994). Picture-name priming in the cerebral hemispheres. Brain and Language, 47, 300–313. Coney, J., & Fitzgerald, J. (2000). Gender differences in the recognition of laterally presented affective nouns. Cognition and Emotion, 14, 325–339. Davidson, R. J. (1992). Anterior cerebral asymmetry and the nature of emotion. Brain and Cognition, 20, 125–151. Davidson, R. J. (1998). Affective style and affective disorders: Perspectives from affective neuroscience. Cognition and Emotion, 12, 307–330. Davidson, R. J., Ekman, P., Saron, C., Senulis, J., & Friesen, W. (1990). Approach-withdrawal and cerebral asymmetry: Emotional expression and brain physiology. Journal of Personality and Social Psychology, 58, 330–341. Delbridge, A. (Ed.). (1987). The Macquarie dictionary (2nd ed.). Sydney: The Macquarie Library. Denny, E. B., & Hunt, R. (1992). Affective valence and memory in depression: Dissociation of recall and fragment completion. Journal of Abnormal Psychology, 101, 575–580. Eviatar, Z., & Zaidel, E. (1991). The effects of word length and emotionality on hemispheric contribution to lexical decision. Neuropsychologia, 29, 415–428. Gainotti, G., Caltagirone, C., & Zoccolotti, P. (1993). Left/right and cortical/subcortical dichotomies in the neuropsychological study of human emotions. Cognition and Emotion, 7, 71–93.

Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behaviour, 23, 553–568. Graves, R., Landis, T., & Goodglass, H. (1981). Laterality and sex differences for visual recognition of emotional and non-emotional words. Neuropsychologia, 19, 95–102. Hartikainen, K., Ogawa, K., & Knight, R. (2000). Transient interference of right hemisphere function due to automatic emotional processing. Neuropsychologia, 38, 1576–1580. Heller, W. (1993). Neuropsychological mechanisms of individual differences in emotion, personality, and arousal. Neuropsychology, 7, 476–489. Hocking, D., & Collins, M. A. (in press). A transfer-appropriate processing account of memory biases for threat in speech anxiety. Cognition and Emotion. Jacoby, L. (1983). Remembering the data: Analysing interactive processes in reading. Journal of Verbal Learning and Verbal Behaviour, 22, 485–508. John, C. H. (1988). Emotionality ratings and free association norms of 240 emotional and non-emotional words. Cognition and Emotion, 2, 49–70. Kinsbourne, M., & Bemporad, B. (1984). Lateralization of emotion: A model and the evidence. In N. A. Fox & R. J. Davidson (Eds.), The psychobiology of affective development (pp. 259–291). Hillsdale, NJ: Erlbaum. Kucera, H., & Francis, W. (1967). Computational analysis of present-day American English. Providence, RJ: Brown University Press. Leshner, G., & Coyle, J. R. (2000). Memory for television news: Match and mismatch between processing and testing. Journal of Broadcasting and Electronic Media, 4, 599–613. Leventhal, G. (1988). Cerebral dominance and attentional bias in word recognition. Perceptual and Motor Skills, 66, 791–800. MacLeod, C. M., & Bassili, J. (1989). Are implicit and explicit tests differentially sensitive to item-specific vs. relational information? In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (pp. 159–172). NJ: Lawrence Erlbaum Associates Inc. Mandal, M., Tandon, S., & Asthana, H. (1991). Right brain damage impairs recognition of negative emotions. Cortex, 27, 247–253. Marsolek, C. J. (1995). Abstract visual-form representation in the left cerebral hemisphere. Journal of Experimental Psychology: Human Perception and Performance, 21, 375–386. Marsolek, C. J. (1999). Dissociable neural subsystems underlie abstract and specific object recognition. Psychological Science, 10, 111– 118. Marsolek, C., & Hudson, T. (1999). Task and stimulus demands influence letter-case-specific priming in the right hemisphere. Laterality, 4, 127–147. Marsolek, C. J., Kosslyn, S., & Squire, L. (1992). Form-specific visual priming in the right cerebral hemisphere. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 492–508. Marsolek, C., Nicholas, C., & Andresen, D. (2002). Interhemispheric communication of abstract and specific visual-form information. Neuropsychologia, 40, 1983–1999. Mulligan, N., Guyer, P., & Belland, A. (1999). The effects of levels-ofprocessing and organization on conceptual implicit memory in the category exemplar production test. Memory and Cognition, 27, 633–647. Nagae, S., & Moscovitch, M. (2002). Cerebral hemispheric differences in memory of emotional and non-emotional words in normal individuals. Neuropsychologia, 40, 1601–1607. Neely, J. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & G. Humphreys (Eds.), Basic processes in reading: Visual word recognition. Hillsdale, NJ: Erlbaum. Paivio, A., Yuille, J., & Madigan, A. (1968). Concreteness, imagery, and meaningfulness values for 925 nouns. Journal of Experimental Psychology: Monograph Supplement, 76, 1–25.

M.A. Collins, A. Cooke / Neuropsychologia 43 (2005) 1529–1545 Peterson, S. E., Fox, P., Posner, M., Mintum, M., & Raichle, M. (1988). Positron emission tomographic studies of the cortical anatomy of single-word processing. Nature, 331, 585–589. Richards, A., French, C., & Dowd, R. (1995). Hemisphere asymmetry and the processing of emotional words in anxiety. Neuropsychologia, 33, 835–841. Riskind, W., & Lane, J. (1987). Emotional states and memory biases: Effects of cognitive priming and mood. Journal of Personality and Social Psychology, 52, 91–99. Roediger, H., & Blaxton, T. (1987). Effects of varying modality, surface features, and retention interval on priming in word-fragment completion. Memory and Cognition, 15, 379–388. Roediger, H., & McDermott, K. (1993). Implicit memory in normal human participants. In F. Boller, & J. Grafman (Eds.), Journal of Neuropsychology, 8, 63–131. Roediger, H., & Srinivas, K. (1993). Specificity of operations in perceptual priming. In P. Graf & M. Masson (Eds.), Implicit memory: New directions in cognition, development, and neuropsychology (pp. 17–48). Hillsdale, NJ: Erlbaum Associates. Roediger, H., Weldon, M., & Challis, B. (1989). Explaining dissociations between implicit and explicit measures of retention: A processing account. In H. Roediger & F. L. Craik (Eds.), Varieties of memory and consciousness: Essays in honour of Endel Tulving (pp. 3–14). Hillsdale, NJ: Erlbaum Associates. Roget, P. (1972). Roget’s thesaurus of synonyms and antonyms. London: Galley Press. Rubin, D., & Friendly, M. (1986). Predicting which words get recalled: Measures of free recall, availability, goodness, emotionality, and pronounciability for 925 nouns. Memory and Cognition, 14, 79– 94. Schacter, D. L. (1992). Priming and multiple memory systems: Perceptual mechanisms of implicit memory. Journal of Cognitive Neuroscience, 4, 244–256.

1545

Schmitt, J. J., Hartje, W., & Willmes, K. (1997). Hemispheric asymmetry in the recognition of emotional attitude conveyed by facial expression, prosody, and propositional speech. Cortex, 33, 65–81. Shapiro, S. I., & Palermo, D. S. (1968). An atlas of normative free association data. Psychonomic Monograph Supplements, 2, 219–250. Strauss, E. (1983). Perception of emotional words. Neuropsychologia, 21, 99–103. Summers, A. (1996). The representation of emotion in the cerebral hemispheres. Unpublished honours thesis, Murdoch University, Perth, Western Australia. Taylor, S. E. (1991). Asymmetrical effects of positive and negative events: The mobilization–minimization hypothesis. Psychological Bulletin, 110, 67–85. Thomson, D., Meredith, S., & Browning, C. (1976). Monash free association norms. Department of Psychology, Monash University. Toglia, M., & Battig, W. (1978). Handbook of semantic word norms. New Jersey: Lawrence Erlbaum Associates. Vaidya, C., Gabrieli, J., Keane, M., Monti, L., Gutuerrez-Rivas, H., & Zarella, M. (1997). Evidence for multiple mechanisms of conceptual priming in implicit memory tests. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1324–1343. Van Strien, J., & Morpurgo, M. (1992). Opposite hemispheric activations as a result of emotionally threatening and non-threatening words. Neuropsychologia, 30, 845–848. Watkins, P., Martin, C., & Stern, L. (2000). Unconscious memory bias in depression: Perceptual and conceptual processes. Journal of Abnormal Psychology, 109, 282–289. Williams, J. M., & McDowell, J. (2001). Implicit memory and depression: An analysis of perceptual and conceptual processes. Cognition and Emotion, 15, 803–812. Williams, J. M., Watts, F., MacLeod, C. M., & Mathews, A. (1997). Cognitive psychology and emotional disorders (2nd ed.). Chichester: Wiley.