Consciousness and Cognition EDITOR William P. Banks Pomona College, Claremont, California
Bruce Bridgeman University of California Santa Cruz
ASSOCIATE EDITORS Axel N. Cleeremans
James T. Enns
Universite´ Libre de Bruxelles
University of British Columbia
Antti Revonsuo University of Turku
EDITORIAL BOARD Jackie Andrade
Steven A. Hillyard
Keith Oatley
University of Plymouth
University of California San Diego
Ontario Institute for Studies in Education
J. Allan Hobson
Steven Palmer
Bernard J. Baars The Neurosciences Institute San Diego
Talis Bachmann
Massachusetts Mental Health Center
University of Tartu
Larry L. Jacoby
Alan Baddeley
New York University
MRC Applied Psychology Unit
E. Roy John
John Bargh
New York University Medical Center
New York University
Arthur L. Blumenthal
John F. Kihlstrom
The New School University
University of California Berkeley
Gordon H. Bower
Christof Koch
Stanford University
Deborah Burke Pomona College
Wallace Chafe University of California Santa Barbara
David Chalmers Antonio Damasio
Stephen LaBerge
University of Iowa
Stanford University
Meredyth Daneman
Donald G. MacKay
University of Wisconsin
Daniel C. Dennett
University of Louisville
Ernst Pçppel Ludwig-MaximiliansUniversitt, Mnchen
William Prinzmetal University of California Berkeley
Arthur Reber
David Rosenthal
David LaBerge
University of Arizona Tucson
Richard Davidson
John Pani
California Institute of Technology Brooklyn College of CUNY Stephen M. Kosslyn Harvard University Eyal Reingold University of Toronto Alfred B. Kristofferson Ontario, Canada University of California Irvine
University of Toronto
University of California Berkeley
University of California Los Angeles
George Mandler University of California San Diego
Tufts University
Bruce Mangan
Andreas K. Engel Hamburg University
University of California Berkeley
Matthew Erdelyi
Graduate School of CUNY
Daniel Schacter Harvard University
Arnold Scheibel University of California Los Angeles
Jonathan W. Schooler University of Pittsburgh
Tim Shallice University College London
Jerome L. Singer
Anthony Marcel
Yale University
Brooklyn College of CUNY
MRC Applied Psychology Unit
David Spiegel
Owen Flanagan
Hazel R. Markus
Duke University
University of Michigan
Stanford University School of Medicine
David Galin
Philip M. Merikle
Petra Stoerig
Langley Porter Psychiatric Institute, San Francisco
University of Waterloo
Heinrich-Heine-Universitat
Thomas Metzinger
Giulio Tononi
Michael S. Gazzaniga
The Neurosciences Institute
Dartmouth College
Johannes GutenbergUniversitt Mainz
Anthony G. Greenwald
Jeff Miller
University of Nottingham
University of Washington
University of Otago
Henk J. Haarman
Michael C. Mozer
University of Maryland
University of Colorado
Stevan Harnad
W. Trammell Neill
Princeton University
University at Albany
Geoffrey Underwood Daniel M. Wegner Harvard University
Charles Yingling University of California San Francisco
Consciousness and Cognition 19 (2010) 505–519
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Tactile expectations and the perception of self-touch: An investigation using the rubber hand paradigm Rebekah C. White a,*, Anne M. Aimola Davies a,b,c,d, Terri J. Halleen d, Martin Davies a,b a
Department of Experimental Psychology, University of Oxford, United Kingdom Faculty of Philosophy, University of Oxford, United Kingdom c NIHR Biomedical Research Centre, University of Oxford, United Kingdom d Department of Psychology, The Australian National University, Australia b
a r t i c l e
i n f o
Article history: Received 12 March 2009 Available online 24 February 2010 Keywords: Body illusion Expectation violation Self-touch Sensory Tactile
a b s t r a c t The rubber hand paradigm is used to create the illusion of self-touch, by having the participant administer stimulation to a prosthetic hand while the Examiner, with an identical stimulus (index finger, paintbrush or stick), administers stimulation to the participant’s hand. With synchronous stimulation, participants experience the compelling illusion that they are touching their own hand. In the current study, the robustness of this illusion was assessed using incongruent stimuli. The participant used the index finger of the right hand to administer stimulation to a prosthetic hand while the Examiner used a paintbrush to administer stimulation to the participant’s left hand. The results indicate that this violation of tactile expectations does not diminish the illusion of self-touch. Participants experienced the illusion despite the use of incongruent stimuli, both when vision was precluded and when visual feedback provided clear evidence of the tactile mismatch. Ó 2009 Elsevier Inc. All rights reserved.
1. Introduction How do you know that the hand you see before you is your own? The systematic study of body ownership poses a challenge to cognitive psychologists for the simple reason that the body is always there (James, 1890), existing as ‘‘a backdrop to whatever one is thinking, experiencing, or doing, though its various parts are not being monitored” (Kinsbourne, 1995, p. 217). Over the past decade, the rubber hand paradigm (Botvinick & Cohen, 1998) has provided an invaluable tool for experimental investigations into body ownership. With this paradigm, individuals experience a sense of ownership over a prosthetic limb. Researchers have used this illusion of ownership to understand body awareness better. They manipulate factors thought to underlie self-attribution of body parts, and they assess the impact of these manipulations on the sense of ownership of the prosthetic limb. Candidate factors which may play a role in self-attribution of body parts include: visual signals, somatosensory signals, proprioceptive signals and higher-order representations of the body. In the visual rubber hand paradigm (Botvinick & Cohen, 1998), the participant views a prosthetic hand being stimulated while her own hand – hidden from view – receives synchronous stimulation. The participant may feel as if the touch on her own hand is occurring at the location where she sees the prosthetic hand being touched. The participant may also experience the illusion that the prosthetic hand is her own hand. The rubber hand illusion is most often assessed using questionnaires that measure the participant’s subjective experience (Botvinick & Cohen, 1998; Mussap & Salton, 2006; Pavani, Spence, & Driver, 2000; Schaefer, Flor, Heinze, & Rotte, 2007). The illusion can also be assessed by measuring: (1) proprioceptive drift * Corresponding author. Address: Department of Experimental Psychology, University of Oxford, South Parks Road, OX1 3UD, United Kingdom. Fax: +44 (0) 1865 310447. E-mail address:
[email protected] (R.C. White). 1053-8100/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2009.08.003
506
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
– the participant may mislocalise her own hidden hand, indicating that the felt position is closer to the observed prosthetic hand (Botvinick & Cohen, 1998; Costantini & Haggard, 2007; Durgin, Evans, Dunphy, Klostermann, & Simmons, 2007; Holmes, Crozier, & Spence, 2004; Tsakiris & Haggard, 2005; Tsakiris, Hesse, Boy, Haggard, & Fink, 2007; Tsakiris, Prabhu, & Haggard, 2006); (2) skin temperature – the participant may undergo a drop in the temperature of her own hidden hand because taking ownership of the observed prosthetic hand disrupts ownership of her own hand (Moseley, Olthof, Venema, Don et al., 2008); and (3) galvanic skin response – the participant may show an increased galvanic skin response if the observed prosthetic hand is threatened (e.g., by a needle) because it has been assimilated into the participant’s body image (Armel & Ramachandran, 2003; Hägni, Eng, Hepp-Reymond, Holper et al., 2008). The following conditions are best for demonstrating the visual rubber hand illusion: (1) stimulation on the participant’s hand is synchronous with stimulation on the observed prosthetic hand (Armel & Ramachandran, 2003; Botvinick & Cohen, 1998; Shimada, Fukuda, & Hiraki, 2009); (2) orientation of the strokes on the participant’s hand is matched to those on the observed prosthetic hand (Costantini & Haggard, 2007); and (3) postural alignment of the participant’s hand is matched to that of the observed prosthetic hand (Costantini & Haggard, 2007; Ehrsson, Spence, & Passingham, 2004; Pavani et al., 2000; Tsakiris & Haggard, 2005), with the observed prosthetic hand positioned (3a) in a biologically-plausible location relative to the participant’s own body (Armel & Ramachandran, 2003) and (3b) close to the participant’s own hidden hand (Lloyd, 2007). It may also be important for the observed prosthetic hand to correspond to the participant’s pre-existing higher-order body representation (Armel & Ramachandran, 2003; Costantini & Haggard, 2007; Haans, IJsselsteijn, & de Kort, 2008; Pavani & Zampini, 2007; Tsakiris et al., 2007). However, the research literature is divided on this point. Armel and Ramachandran (2003) found that the rubber hand illusion was elicited despite visual inconsistencies between the participant’s own hand and the prosthetic hand, including skin tone, hand size and distinguishing visual features such as nail polish (see also Longo, Schüür, Kammers, Tsakiris, & Haggard, 2009). Moreover, they reported that the illusion could even be elicited in the absence of a prosthetic hand, insofar as participants ‘‘often reported sensations arising from the table surface” (p. 1499) when it was stroked and tapped in precise synchrony with the stimulation administered to the participant’s hidden hand. This finding was not supported by Haans and colleagues (2008), who found that the subjective experience of the rubber hand illusion (as measured by questionnaire items) was significantly diminished when the participant viewed the table (rather than a prosthetic hand) being stimulated. Likewise, Tsakiris and Haggard (2005) found that synchronous stimulation of a wooden stick and the participant’s hand did not produce the illusion. Tsakiris and colleagues (2007) have also shown that the illusion is not elicited if the participant is observing stimulation on a prosthetic left hand while her own (hidden) right hand is stimulated. How do we account for these findings that the rubber hand illusion may be diminished when the participant views stimulation on a non-hand or wrong-hand object? One possible explanation is that it may be more difficult for the participant to experience illusory ownership over an object that does not correspond to her higher-order body representation, simply because the object conflicts with her stored representation of what belongs to her own body or to the human body more generally. An alternative explanation is that the participant may expect the stimulation on her own hand to feel a certain way and this expectation may depend on the viewed object. If the viewed object has a non-skin texture, stimulation on the participant’s own skin may feel different from what she would expect given the texture of the viewed object (Armel & Ramachandran, 2003; Haans et al., 2008) and this violation of tactile expectations may be sufficient to break the illusion. These alternative views are best conceptualised using cases from the literature. Armel and Ramachandran described four participants (out of 120) with particularly hairy hands who spontaneously indicated that ‘‘the illusion was ruined when their hand was touched in areas of high hair density” (p. 1504). The researchers suggested that it was ‘‘a mismatch in the expected (from visual information) versus felt type of touch, rather than just the visual inconsistencies of hair versus no hair, that diminished the illusion” (p. 1504). Not dissimilarly, Haans and colleagues (2008) found that, when the illusion was evoked using a prosthetic hand wearing a latex glove rather than a prosthetic hand of natural skin texture, ‘‘several participants remarked that their tactile sensations did not match those generally perceived while wearing gloves” (p. 393). But, as noted by Haans and colleagues, visual similarity and tactile expectation are confounded in these examples – the sensations on the participant’s hand match expectations when viewing the visually similar object but not when viewing the visually dissimilar object. Hence, it is not clear whether the illusion is affected by use of a visually dissimilar hand that the participant is not able to incorporate into a higher-order body representation, or by the violation of tactile expectation. Schütz-Bosbach, Tausche, and Weiss (2009) have very recently isolated the role of tactile expectation, thus addressing the concern that higher-order body representation and tactile expectation are confounded in previous research. Rather than manipulating the match between tactile properties of the observed prosthetic hand and of the participant’s own hand, the researchers manipulated the match between tactile properties of the stimulus that was administered to the observed prosthetic hand and of the stimulus administered to the participant’s own hand. For example, the participant viewed a prosthetic hand being touched with a soft fabric while receiving stimulation from a rough fabric on her own (hidden) hand. In this case there was no confound. The participant’s tactile expectation was violated but the participant was viewing stimulation upon a realistic prosthetic hand that could be incorporated into her higher-order body representation. Participants experienced the rubber hand illusion even when their tactile expectations were violated. When considered together with previous studies demonstrating that the rubber hand illusion is diminished if the participant views stimulation on a nonhand or wrong-hand object (or a hand that is markedly dissimilar to the participant’s own hairy hand; Armel & Ramachandran, 2003), Schütz-Bosbach and colleagues’ finding suggests that the disruption to the illusion occurs because the viewed
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
507
object cannot be incorporated into the participant’s higher-order body representation, rather than because the participant’s tactile expectations are violated. Here we propose that the interplay of higher-order body representation and tactile expectation may also be investigated using the non-visual rubber hand paradigm (Ehrsson, Holmes, & Passingham, 2005). With vision precluded, the Examiner guides the participant’s hand in administering stimulation to a prosthetic hand while the Examiner administers synchronous stimulation to the participant’s other hand. The participant may experience the illusion that she is touching her own hand, even when the two hands are separated by 15 cm. The illusion of self-touch can be assessed using questionnaires as well as measures of proprioceptive drift. As with the visual rubber hand illusion, a participant who experiences the non-visual rubber hand illusion, that is, the self-touch illusion, may mislocalise her ‘touched’ hand, indicating that its felt position is closer to the prosthetic hand than is actually the case (Ehrsson et al., 2005). In the elegant study by Ehrsson and colleagues (2005), the ‘‘participants, the experimenter, and the rubber hand all wore identical plastic surgical gloves to make the tactile surfaces of the two hands [participant and prosthetic] as similar as possible to each other” (p. 10565). When the participant’s stimulation of the prosthetic hand and the Examiner’s stimulation of the participant’s hand were synchronous, the participant experienced the compelling illusion of touching her own hand. In a control condition, the Examiner guided the participant instead to tap the bristles of a small dish brush with one hand while the Examiner synchronously stimulated the participant’s other hand. In this condition, ‘‘no illusion of self-touch was typically elicited” (p. 10566), but it is not clear whether this was due to the violation of higher-order body representation or the violation of tactile expectation since these factors were confounded. Higher-order body representation was violated because a dish brush cannot be assimilated into one’s body image, and tactile expectations were violated because under conditions of self-touch the participant would expect her administering index finger to feel a skin-like surface rather than the bristles of a brush. In the current study, our aim is to investigate the role of tactile expectation in the self-touch illusion, without confounding higher-order body representation and tactile expectation. We manipulate tactile properties of the administering stimulus (as in Schütz-Bosbach et al., 2009) rather than tactile properties of the object to which the participant administers stimulation. If the illusion of self-touch requires a match between expected and felt sensations, the participant will not experience the illusion when she is touching the prosthetic hand with one stimulus and receiving stimulation from a different stimulus. Alternatively, if expectations about tactile sensations do not affect the illusion, the impression of self-touch should persist provided there is synchronous stimulation of the two hands. 1.1. Overview of experiments Three experiments are presented. In Experiment 1, we assessed whether the illusion of self-touch could be elicited when participants used instruments, such as a paintbrush or a stick, as well as when they used an index finger to administer touch. In the only previous experiment using the non-visual rubber hand paradigm (Ehrsson et al., 2005), stimulation was always administered with the index finger. Experiment 1 was therefore a necessary precursor to later experiments, which introduced incongruent stimulation. To examine whether the non-visual rubber hand illusion breaks down under conditions of incongruent stimulation, it was first necessary to show that the illusion could be elicited using a variety of stimuli administered under congruent conditions and with vision precluded. In Experiment 2, we assessed whether the illusion of self-touch could be elicited when the participant administered stimulation to the prosthetic hand using her index finger while her other hand was touched with – an incongruent stimulus – a paintbrush. This tactile mismatch has an effect on both of the participant’s hands: the participant’s administering hand is receiving tactile sensations consistent with touching a prosthetic hand while the other hand is receiving tactile sensations consistent with being touched with a paintbrush. This experiment was conducted with vision precluded. In Experiment 3, we combined the methodologies of the traditional visual (Botvinick & Cohen, 1998) and non-visual (Ehrsson et al., 2005) rubber hand paradigms. The participants were involved in the administration of touch as they are in the non-visual paradigm but the paradigm was conducted with vision permitted. Stimulation was either congruent or incongruent. In the congruent condition, the participant administered stimulation to the prosthetic hand using her index finger while her other hand was touched with – a congruent stimulus – the Examiner’s index finger. In the incongruent condition, the participant administered stimulation to the prosthetic hand using her index finger while her other hand was touched with – an incongruent stimulus – a paintbrush. On a given trial, the participant had vision either of her right hand administering stimulation to the prosthetic hand or of her left hand receiving stimulation from the Examiner. 2. Experiment 1 2.1. Method 2.1.1. Participants Thirty-six right-handed (Oldfield, 1971) participants (aged 19–33 years; 19 females) took part in Experiment 1. Participants gave informed consent and the study was approved by the University of Oxford Research Ethics Committee and conducted in accordance with the ethical standards laid down in the 2008 Declaration of Helsinki.
508
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
2.1.2. Materials and procedure The participant was seated at a testing table with a prosthetic left hand positioned at body midline. The participant’s own left hand was positioned to the left of the prosthetic hand with the two index fingers 15 cm apart. The Examiner sat on the opposite side of the table to the participant. The Examiner guided the participant’s right hand in administering stimulation to the index finger of the prosthetic hand, using (a) the participant’s index finger, (b) a paintbrush or (c) a stick. The paintbrush and the stick were mounted on identical handles. The participant was instructed to keep her eyes closed throughout the experimental trial, and to relax and allow her right hand to be guided by the Examiner. At the same time, the Examiner administered stimulation to the index finger of the participant’s left hand using the same (congruent) stimulus as was used to stimulate the prosthetic hand. For example, the participant was guided to administer stimulation to the prosthetic hand using a paintbrush while the Examiner used an identical paintbrush to administer stimulation to the participant’s left hand (Fig. 1). Stimulation comprised strokes and taps administered to the proximal phalanx of the index finger (including the metacarpal phalangeal joint and the proximal interphalangeal joint). Strokes were always unidirectional towards the finger tip. The Examiner’s stimulation of the participant’s left hand was timed to be either (a) synchronous or (b) asynchronous with the stimulation that the participant administered to the prosthetic hand. Each trial lasted for 60 s, and the order of conditions (i.e., finger, paintbrush, stick) was counterbalanced across participants using a Latin-Square design. 2.1.3. Measure of proprioceptive drift To examine the extent of proprioceptive drift associated with the self-touch illusion, we adapted the calibration method used by Ehrsson and colleagues (2005). Three measurements were obtained: two pre-stimulation measurements and one post-stimulation measurement. The pre-stimulation measurements were averaged to give a baseline measure of felt position. Proprioceptive drift was calculated as the change in felt position of the index finger from baseline. The prosthetic hand was removed from the table when these measurements were taken, immediately before and after each 60-s trial. The participant used her right index finger to point to the felt position of her left index finger. The participant extended her right arm at 45° to the right of the body’s midsagittal plane, and slid her right index finger along the testing table until it was in line with the felt position of her left index finger. Participants were instructed to point quickly using a single movement. The Examiner recorded the distance between the participant’s indicated and actual finger position. Note
Fig. 1. Experimental set-up for stimulation administered using the paintbrush. Note that the index finger of the participant’s left hand is positioned 15 cm from the index finger of the prosthetic hand.
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
509
that the participant closed her eyes prior to the two pre-stimulation measurements, and was asked to maintain eyes closed for the duration of the experimental trial, that is, during the stimulation period and until the post-stimulation measure of felt position was obtained. 2.1.4. Rubber hand illusion questionnaire Following each trial, the participant indicated her level of agreement with two statements: (1) It felt as if I were touching my own hand. (2) It felt as if my left hand were shrinking. The first statement was adapted from Ehrsson and colleagues (2005) and was used to measure the participant’s subjective experience of the rubber hand illusion. This is the only statement used previously in the research literature to quantify the non-visual rubber hand illusion. The second statement served as a control for suggestibility, that is, a statement about an experience that the paradigm was not designed to evoke. A seven-point visual analogue scale (0 = not at all; 6 = very strongly agree) was used to rate agreement with the statements.1 2.2. Results 2.2.1. Proprioceptive drift We performed a mixed between- and within-subjects analysis of variance (ANOVA) on proprioceptive drift, with one between-subjects factor and two within-subjects factors. The between-subjects factor was order of the stimuli used to administer stimulation (order one: finger, paintbrush, stick; order two: paintbrush, stick, finger; order three: stick, finger, paintbrush). The within-subjects factors were mode of stroking (synchronous, asynchronous) and the stimulus used (finger, paintbrush, stick). There was a main effect of administration order (F(2, 33) = 5.336, p = .01, multivariate partial eta-squared = .244). Post-hoc comparisons using the Scheffé test indicated significantly greater proprioceptive drift (p = .014) for participants exposed to order two (paintbrush, stick, finger: M = 2.281 cm) compared with order one (finger, paintbrush, stick: M = .500 cm). There was a main effect of mode of stroking (F(1, 33) = 7.922, p = .008, multivariate partial eta-squared = .194), with greater proprioceptive drift for synchronous (M = 1.829 cm) compared with asynchronous (M = .648 cm) stimulation. There was no main effect of stimulus (F(2, 32) = 1.369, p = .269, multivariate partial eta-squared = .079), and only one interaction (administration order and stimulus) approached significance (p = .091; all other p values >.190). 2.2.2. Subjective experience of the rubber hand illusion We performed a mixed between- and within-subjects ANOVA on the participant’s subjective experience of the rubber hand illusion, with one between-subjects factor (administration order) and two within-subjects factors (mode of stroking and stimulus). There was no main effect of administration order (F(2, 33) = .178, p = .838, multivariate partial etasquared = .011). There was a main effect of mode of stroking (F(1, 33) = 88.743, p < .001, multivariate partial etasquared = .729), with higher illusion ratings for synchronous (M = 3.106) compared with asynchronous (M = .602) stimulation. There was also a main effect of stimulus (F(2, 32) = 8.237, p = .001, multivariate partial eta-squared = .340). There were significant two-way interactions of administration order and stimulus (F(4, 64) = 2.807, p = .033, multivariate partial etasquared = .149) and mode of stroking and stimulus (F(2, 32) = 14.170, p < .001, multivariate partial eta-squared = .470). There was also a significant three-way interaction of administration order, mode of stroking and stimulus (F(4, 64) = 2.891, p = .029, multivariate partial eta-squared = .153). Separate ANOVA were conducted for synchronous and asynchronous stimulation to investigate these interactions. For synchronous stimulation (Fig. 2: top panel), there was no main effect of administration order (F(2, 33) = .247, p = .783, multivariate partial eta-squared = .015) but there was a main effect of stimulus (F(2, 32) = 13.105, p < .001, multivariate partial eta-squared = .450) and a significant interaction of administration order and stimulus (F(4, 64) = 3.795, p = .008, multivariate partial eta-squared = .192). As depicted in the top panel of Fig. 2, participants in administration order 2 provided equally high illusion ratings for the three stimuli, (F(2, 10) = 1.254, p = .327, multivariate partial eta-squared = .201). In contrast, participants in administration order 1 had a significant difference in the strength of illusion ratings for the three stimuli (F(2, 10) = 6.442, p = .016, multivariate partial eta-squared = .563). Pairwise comparisons (adjusting for multiple comparisons) indicated higher illusion ratings for the paintbrush (M = 3.750) than for the finger (M = 1.833, p = .010). Participants in administration order 3 also had a significant difference in the strength of illusion ratings for the three stimuli (F(2, 1 The strength of the rubber hand illusion is frequently measured using a seven-point scale with negative through positive values. Participants provide an agreement rating with statements probing experience of the illusion (e.g., 3 = strongly disagree; +3 = strongly agree). However, it has been argued recently (Miles, 2008, April) that participants who do not experience the illusion may find it difficult to rate non-agreement (i.e., 1 vs. 2 vs. 3). Therefore we use a scale in which failure to experience the illusion receives a zero rating (not at all) and experience of the illusion is rated 1 (slightly agree) through 6 (very strongly agree). This scale provides greater range for rating relative strength for the vast majority of participants who do experience the illusion. Notably, results from our scale (in the conditions in which stimulation was administered with a finger) are comparable to the pioneering study by Ehrsson et al. (2005). In the current study, 31 of 36 participants (86%) provided ‘agree’ ratings compared with 25 of 32 participants (78%) in the study by Ehrsson et al. (v2(1, N = 68) = .295, p = .587). Recent studies using a similar (positive) scale to ours include: Haans et al. (2008): 0 = not at all; 10 = completely; Moseley et al. (2008): 0 = not at all vivid; 10 = completely vivid.
510
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
Fig. 2. Top panel – synchronous stimulation; Bottom panel – asynchronous stimulation. Illusion ratings as a function of administration order and stimulus. Error bars represent Standard Error of the Mean.
10) = 7.485, p = .010, multivariate partial eta-squared = .600). Pairwise comparisons (adjusting for multiple comparisons) indicated higher illusion ratings for the paintbrush (M = 4.250) than for the stick (M = 2.167, p = .006) or the finger (M = 2.500, p = .014). Thus the interaction of administration order and stimulus indicates that the paintbrush may have a significant effect on the illusion ratings of the finger and stick stimuli if they follow paintbrush in the administration order. For asynchronous stimulation (Fig. 2: bottom panel), there was no main effect of administration order (F(2, 33) = .082, p = .922, multivariate partial eta-squared = .005) or stimulus (F(2, 32) = .283, p = .755, multivariate partial etasquared = .017) and there was no interaction of administration order and stimulus (F(4, 64) = .791, p = .536, multivariate partial eta-squared = .047). 2.2.3. Suggestibility The mean rating for the control statement was less than .5 for each stimulus (finger, paintbrush, stick), regardless of whether stimulation was synchronous or asynchronous. Paired t-tests (using an adjusted alpha level of .0167) confirmed that there was no difference in agreement ratings for synchronous versus asynchronous stimulation: finger (t(35) = .211, p = .834); paintbrush (t(35) = .452, p = .654); stick (t(35) = .298, p = .768).
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
511
A further comparison (using an adjusted alpha level of .0167) confirmed higher agreement ratings for the illusion statement (It felt as if I were touching my own hand) compared with the control statement (It felt as if my left hand were shrinking) in the synchronous stimulation conditions: finger (t(35) =7.032, p < .001); paintbrush (t(35) = 10.736, p < .001); stick (t(35) = 7.289, p < .001). 2.2.4. Correlation of proprioceptive drift and questionnaire ratings For each stimulus, there was a significant correlation between proprioceptive drift and the difference between agreement ratings for the illusion and control statements on the questionnaire: finger (r = .534, p < .001); paintbrush (r = .371, p = .013); stick (r = .287, p = .045).
3. Experiment 2 3.1. Method 3.1.1. Participants Twenty-three new right-handed participants took part in Experiment 2. One participant was excluded due to failure to maintain eyes closed. Data from 22 participants (aged 18–31 years; 16 females) were analysed. 3.1.2. Materials and procedure The procedure was closely modelled on Experiment 1. Experiment 2 comprised four 60-s trials, two with congruent stimulation and two with incongruent stimulation. The Examiner guided the participant to use the index finger of one hand to administer stimulation to the prosthetic hand while the Examiner administered stimulation to the participant’s other hand, either with an index finger (congruent stimulation – Trials 1 and 2) or with a paintbrush (incongruent stimulation – Trials 3 and 4). Stimulation was timed so as to be (a) synchronous (Trials 1 and 3) or (b) asynchronous (Trials 2 and 4). To ensure that the participant was relaxed with the procedure for administering stimulation, the Examiner and participant practised the guided stimulation technique before the experimental trials began. The participant was instructed to keep her eyes closed throughout the experimental trial, and to relax and allow her right hand to be guided by the Examiner. 3.1.3. Measure of proprioceptive drift Proprioceptive drift was assessed with the pointing judgment task used in Experiment 1. As with Experiment 1, the participant closed her eyes prior to the two pre-stimulation measurements, and she was asked to maintain eyes closed for the duration of the experimental trial, that is, during the stimulation period and until the post-stimulation measure of felt position was obtained. 3.1.4. Rubber hand illusion questionnaires Following each trial, the participant completed two short questionnaires. The first questionnaire comprised two statements as in Experiment 1: (1) It felt as if I were touching my own hand. (2) It felt as if my left hand were shrinking. The second questionnaire comprised eight statements designed to measure the participant’s perception of the stimuli used in the experimental trials. The statements were based on the experiences reported by participants who took part in earlier experiments piloting the use of incongruent stimulation: (1) (2) (3) (4) (5) (6) (7) (8)
It felt as if I were administering touch with a It felt as if I were administering touch with a It felt as if I were administering touch with a My left index finger felt as if it were touched My left index finger felt as if it were touched My left index finger felt as if it were touched My right index finger felt like a finger. My right index finger felt like a brush.
finger. brush. finger and a brush simultaneously. by a finger. by a brush. by a finger and a brush simultaneously.
The order of statements was randomised across trials, and a seven-point visual analogue scale (0 = not at all; 6 = very strongly agree) was used to rate agreement with the statements. Once the participant had completed these subjective items assessing felt experience, that is, her perception of the stimuli used, she was asked to name both the stimulus used to administer stimulation to the prosthetic hand and the stimulus used to administer stimulation to her own hand. This question was designed to verify that the participant understood that the stimuli were incongruent on the critical trials.
512
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
3.2. Results 3.2.1. Proprioceptive drift We performed an ANOVA on proprioceptive drift. The two within-subjects factors were congruence of stimulation (congruent, incongruent) and mode of stroking (synchronous, asynchronous). There was no main effect of congruence (F(1, 21) = .020, p = .890, multivariate partial eta-squared = .001). There was a main effect of mode of stroking (F(1, 21) = 16.220, p = .001, multivariate partial eta-squared = .436), with greater proprioceptive drift for synchronous (M = 2.381 cm) compared with asynchronous (M = .034 cm) stimulation. There was no interaction of congruence and mode of stroking (F(1, 21) = .317, p = .579, multivariate partial eta-squared = .015). See Fig. 3. 3.2.2. Subjective experience of the rubber hand illusion We performed an ANOVA on the participant’s subjective experience of the rubber hand illusion. The within-subjects factors were congruence of stimulation (congruent, incongruent) and mode of stroking (synchronous, asynchronous). There was no main effect of congruence (F(1, 21) = 1.634, p = .215, multivariate partial eta-squared = .072). There was a main effect of mode of stroking (F(1, 21) = 36.426, p < .001, multivariate partial eta-squared = .634), with higher overall illusion ratings for synchronous (M = 3.091) compared with asynchronous (M = .795) stimulation. There was no interaction of congruence and mode of stroking (F(1, 21) = .009, p = .926, multivariate partial eta-squared = .000). See Fig. 4. 3.2.3. Suggestibility The mean rating for the control statement was less than .5 for congruent and incongruent stimulation trials, regardless of whether stimulation was synchronous or asynchronous. Paired t-tests (using an adjusted alpha level of .025) confirmed that there was no difference in agreement ratings for synchronous versus asynchronous stimulation for either congruent stimulation (t(21) = .826, p = .418) or incongruent stimulation (t(21) = 1.449, p = .162). A further comparison (using an adjusted alpha level of .025) confirmed higher agreement ratings for the illusion statement (It felt as if I were touching my own hand) compared with the control statement (It felt as if my left hand were shrinking) in each of the synchronous stimulation conditions: congruent stimulation (t(21) = 6.757, p < .001); incongruent stimulation (t(21) = 5.376, p < .001). 3.2.4. Correlation of proprioceptive drift and questionnaire ratings There was no significant correlation found, for either congruent stimulation or incongruent stimulation, between proprioceptive drift and the difference between agreement ratings for the illusion and control statements on the first questionnaire: congruent stimulation (r = .200, p = .373); incongruent stimulation (r = .251, p = .259). 3.2.5. Perception of the stimulus On the congruent trial with synchronous stimulation, 20 of 22 participants correctly reported receiving stimulation from an index finger. On the incongruent trial with synchronous stimulation, all 22 participants correctly reported receiving
Fig. 3. Mean proprioceptive drift for each of the stimulus conditions in Experiment 1 (collapsing across administration order) and Experiment 2. Error bars represent Standard Error of the Mean.
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
513
Fig. 4. Illusion ratings as a function of stimulus congruence and mode of stroking (synchronous versus asynchronous). Error bars represent Standard Error of the Mean.
stimulation from a brush. (Not surprisingly, all participants correctly reported administering stimulation with an index finger on the congruent and incongruent stimulation trials.) Under conditions of incongruent stimulation, the participant may bind the tactile sensations from each hand when she experiences the illusion of self-touch. This may lead either to the perception that the receptive left hand is being simultaneously touched by two stimuli or, alternatively, to the perception that the index finger is brush-like. Two planned analyses (using an adjusted alpha level of .025) were conducted to assess the impact of stimulus incongruence on tactile perception. First, a paired t-test was conducted on ratings provided for the statement ‘My left index finger felt as if it were touched by a finger and a brush simultaneously’. There was a significant difference (t(21) = 2.42, p = .025) in the level of agreement between the congruent (M = .318) and incongruent stimulation trials (M = 1.41), despite the fact that in both trials the participant received touch from one stimulus only. Second, a paired t-test was conducted on ratings provided for the statement ‘My right index finger felt like a brush’. There was a significant difference (t(21) = 4.143, p < .001) in the level of agreement between the congruent (M = .318) and incongruent stimulation trials (M = 1.64), despite the fact that in both trials the participant’s right index finger was administering touch to the prosthetic hand. 4. Experiment 3 4.1. Method 4.1.1. Participants Twelve participants (aged 19–27 years; 9 females) took part in Experiment 3. The participants were recruited from amongst 16 individuals who had experienced a strong (or very strong) illusion of self-touch in any of the stimulus conditions (finger, paintbrush, stick) used in Experiment 1. Participants were tested at least 1 week following participation in Experiment 1. 4.1.2. Experiment overview Experiment 3 comprised six trials. The baseline conditions (Trials 1 and 4) were conducted with vision precluded, and the remaining trials (Trials 2, 3, 5, 6) with vision permitted. 4.2. Baseline conditions 4.2.1. Materials and procedure Trials 1 and 4 were congruent (finger–finger) 60-s stimulation trials, conducted with vision precluded. The Examiner guided the participant’s right index finger to administer stimulation to the prosthetic hand while the Examiner administered synchronous stimulation to the participant’s left hand. The procedure for stimulation was as in Experiments 1 and 2 except that only strokes of the proximal phalanx of the index finger were used, rather than strokes and taps, and the strokes were always unidirectional towards the wrist, rather than towards the finger tip.
514
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
4.2.2. Rubber hand illusion questionnaire Following each trial, the participant indicated her level of agreement with two statements as in Experiment 1: (1) It felt as if I were touching my own hand. (2) It felt as if my left hand were shrinking. A seven-point visual analogue scale (0 = not at all; 6 = very strongly agree) was used to rate agreement with the statements. 4.3. Vision conditions 4.3.1. Materials and procedure Trials 2, 3, 5 and 6 were conducted with vision permitted. Two examiners were required to conduct these trials. Examiner 1 sat to the left of the participant and Examiner 2 was positioned to the right. Trials 2 and 5 comprised congruent stimulation: Examiner 1 used her right index finger to administer stimulation to the participant’s left hand while Examiner 2 guided the participant’s right index finger to administer stimulation to the prosthetic hand (Fig. 5). Trials 3 and 6 comprised incongruent stimulation: Examiner 1 used a paintbrush to administer stimulation to the participant’s left hand while Examiner 2 guided the participant’s right index finger to administer stimulation to the prosthetic hand. Timing was guided by the stimulation administered by Examiner 1. Six participants were permitted vision of their left hand on Trials 2 and 3 (view-left condition) and vision of their right hand on Trials 5 and 6 (view-right condition). The remaining six participants were permitted vision of their right hand on Trials 2 and 3 (view-right condition) and vision of their left hand on Trials 5 and 6 (view-left condition). In the view-left condition, the participant had vision of her own left hand being stroked by Examiner 1. In the view-right condition, the participant had vision of her own right hand as it was guided by Examiner 2 to stroke the prosthetic hand. A partition was placed between the participant’s left hand and the prosthetic hand to ensure that the participant could only see the hand(s) she was permitted to view. Once Examiner 1 and Examiner 2 had achieved synchronous stimulation, as determined by the participant and Examiner 2, the stopwatch was activated. Sixty seconds of synchronous stimulation followed. 4.3.2. Rubber hand illusion questionnaire Following each trial, the participant indicated her level of agreement with three statements: (1) It felt as if I were touching my own hand. (2) It seemed as if I were observing my right hand stroking my left hand. (3) It felt as if my left hand were shrinking. The first statement was used to measure the participant’s subjective experience of the rubber hand illusion. The second statement was used to measure whether participants experienced ownership of the Examiner’s administering hand (viewleft condition) or the prosthetic hand (view-right condition). The third statement served as a control for suggestibility. A seven-point visual analogue scale (0 = not at all; 6 = very strongly agree) was used to rate agreement with the statements.
Fig. 5. Experimental set-up in a congruent stimulation trial in which the participant was permitted vision of her own left hand as it was touched by Examiner 1. Black fabric was draped over the arms on the viewed side to hide irrelevant cues from clothing (view-left condition) or from the stump of the prosthetic hand (view-right condition).
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
515
4.4. Results Experiment 2 demonstrated that there was no difference in illusion ratings between the congruent and incongruent stimulation conditions. Since we predicted similar non-significant findings for Experiment 3 (see also Schütz-Bosbach et al., 2009), we used an adjusted alpha level of .10 to control for the possibility that a non-significant result may be due to insufficient power (Stevens, 1996). 4.4.1. Vision and the illusion of self-touch To assess the effect of vision of the left (touched) or right (touching) hand, on the illusion that one is touching one’s own hand, we conducted a mixed between- and within-subjects ANOVA on the illusion ratings (Statement 1 ‘It felt as if I were touching my own hand’) for conditions of congruent stimulation (Trials 1, 2, 4, 5). There was one between-subjects factor and two within-subjects factors. The between-subjects factor was order – participants viewed their left hand or their right hand first. The within-subjects factors were vision (vision precluded, vision permitted) and hand viewed by the participant (view-left, view-right). There was no main effect of order (F(1, 10) = .049, p = .830, multivariate partial eta-squared = .005). There was a main effect of vision (F(1, 10) = 5.090, p = .048, multivariate partial eta-squared = .337), with higher illusion ratings for trials conducted with vision precluded (M = 3.271) compared to trials with vision permitted (M = 2.729). There was no main effect of viewed-hand (F(1, 10) = .057, p = .816, multivariate partial eta-squared = .006) but there was a significant interaction of vision and viewed-hand (F(1, 10) = 4.447, p = .061, multivariate partial eta-squared = .308) at the adjusted alpha level of .10. To assess this interaction, we conducted two separate paired t-tests (using an adjusted alpha level of .05). The first t-test demonstrated no difference in illusion ratings between the non-visual trial before participants looked at their left hand (M = 3.04) and the visual trial in which participants looked at their left hand (M = 3.04) (t(11) = .000, p = 1.000). The second t-test demonstrated a difference in illusion ratings between the non-visual trial before participants looked at their right hand (M = 3.5) and the visual trial in which participants looked at their right hand (M = 2.41) (t(11) = 2.493, p = .03). There were no other interactions (all p values >.149). 4.4.2. Stimulus incongruence and the illusion of self-touch To assess whether stimulus incongruence affects the felt illusion that one is touching one’s own hand under conditions permitting vision, we conducted a mixed between- and within-subjects ANOVA on illusion ratings (Statement 1 ‘It felt as if I were touching my own hand’). There was one between-subjects factor and two within-subjects factors. The between-subjects factor was order – participants viewed their left or their right hand first. The within-subjects factors were congruence of stimulation (congruent, incongruent) and hand viewed by the participant (view-left, view-right). There were no main effects of order (F(1, 10) = .000, p = .987, multivariate partial eta-squared = .000), congruence (F(1, 10) = .254, p = .625, multivariate partial eta-squared = .025) or viewed-hand (F(1, 10) = 1.549, p = .242, multivariate partial eta-squared = .134), and there were no significant interactions (all p values >.199). 4.4.3. Stimulus incongruence and the illusion of ownership To assess whether stimulus incongruence affects the illusion of ownership of the prosthetic hand or of the Examiner’s administering hand, we conducted a mixed between- and within-subjects ANOVA on ownership ratings (Statement 2 ‘It seemed as if I were observing my right hand stroking my left hand’). There was one between-subjects factor and two within-subjects factors. The between-subjects factor was order – participants viewed their left or their right hand first. The within-subjects factors were congruence of stimulation (congruent, incongruent) and hand viewed by the participant (view-left, view-right). There was no main effect of order (F(1, 10) = .561, p = .471, multivariate partial eta-squared = .053). There was a main effect of congruence (F(1, 10) = 7.840, p = .019, multivariate partial eta-squared = .439), indicating higher levels of agreement when stimulation was congruent (M = 2.062) compared with when stimulation was incongruent (M = 1.479). There was no main effect of viewed-hand (F(1, 10) = 1.257, p = .289, multivariate partial eta-squared = .112), and there were no significant interactions (all p values >.124). 4.4.4. Suggestibility The mean rating for the control statement was less than .5 for all trials. Paired t-tests (using an adjusted alpha level of .025) confirmed that, in all conditions, participants indicated higher levels of agreement with the illusion statement (It felt as if I were touching my own hand) compared with the control statement (It felt as if my left hand were shrinking): view-left congruent (t(11) = 3.959, p = .002); view-left incongruent (t(11) = 3.006, p = .012); view-right congruent (t(11) = 3.694, p = .004); view-right incongruent (t(11) = 3.764, p = .003).
5. General discussion We set out to investigate whether the illusion of self-touch is affected by violations of tactile expectation. We used the non-visual rubber hand paradigm devised by Ehrsson and colleagues (2005) to address this question. In this paradigm, the Examiner guides the participant to administer stimulation to a prosthetic hand while the Examiner administers synchronous stimulation to the participant’s hand. If the participant experiences the illusion, she reports feeling as if she is touching
516
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
her own hand. Three experiments were conducted. In Experiment 1, we assessed whether the illusion of self-touch could be elicited when participants used instruments, such as a paintbrush or a stick, as well as when they used an index finger to administer touch. In the only previous experiment using the non-visual rubber hand paradigm (Ehrsson et al., 2005), stimulation was always administered with the index finger. In Experiment 2, we assessed whether the illusion of self-touch could be elicited when the participant administered stimulation to the prosthetic hand using her right index finger while her left hand was touched with – an incongruent stimulus – a paintbrush. In Experiment 3, we assessed whether the illusion of self-touch would be diminished when the participant had visual feedback indicating incongruent stimulation. This experiment combined the methodologies of the traditional visual (Botvinick & Cohen, 1998) and non-visual (Ehrsson et al., 2005) rubber hand paradigms. The participants were involved in the administration of touch as they are in the non-visual paradigm but the paradigm was conducted with vision permitted. The primary objective of Experiment 1 was to investigate whether the illusion of self-touch can be elicited when the participant uses an instrument (i.e., paintbrush or stick) instead of an index finger to administer touch to the prosthetic hand while the Examiner uses an identical instrument to administer touch to the participant’s other hand, that is, the hand not involved in administration. The results, showing that synchronous stimulation produced greater proprioceptive drift and higher illusion ratings on the rubber hand questionnaire than asynchronous stimulation, indicate that the self-touch illusion can be elicited when touch is administered with instruments. We also found that order of administration was important to the magnitude of the illusion, so that participants who first received stimulation with the paintbrush (administration order 2) displayed high illusion ratings for all three stimuli and overall greater proprioceptive drift. We suggest that the illusion of self-touch is most readily evoked using the paintbrush, possibly because the flexible nature of the stimulus makes it less likely to be affected by small differences in timing, pressure and pattern of stimulation to the two hands. Once a participant has experienced the illusion with the paintbrush, she may be more prepared to experience it with subsequent stimuli. It may otherwise take time for the participant to be relaxed about the experimental procedure and to allow her hand to be guided by the Examiner. Potential problems were addressed in Experiment 2 by having the participant and Examiner practise the guided stimulation technique before the experimental trials and in Experiment 3, in which two Examiners were involved in administering stimulation, by starting the stopwatch only when the participant and Examiner 2 agreed that synchronous stimulation of the participant’s and the prosthetic hand had been achieved. Having demonstrated that the illusion of self-touch can be elicited using a variety of stimuli in the non-visual rubber hand paradigm, we designed Experiment 2 to assess whether the illusion persists if there is a mismatch in the stimuli used to administer stimulation. Incongruence was set up by having the participant administer touch to the prosthetic hand using her right index finger while the Examiner administered touch to the participant’s left hand using a paintbrush. The results, showing that synchronous stimulation produced greater proprioceptive drift and higher illusion ratings on the rubber hand questionnaire than asynchronous stimulation, indicate that the illusion of touching one’s own hand persists, even when the participant is aware that she is administering touch with a different type of stimulus from the one touching her other hand.2 There was no significant difference in proprioceptive drift when the participant administered touch with a finger and was synchronously touched with a finger (congruent stimulation trial: M = 2.443 cm) compared with when the participant administered touch with a finger and was synchronously touched with a paintbrush (incongruent stimulation trial: M = 2.318 cm). This finding indicates that participants experience mislocalisation in the felt position of the receptive left hand, whether stimulation is congruent or incongruent. Similarly, in response to the statements in the questionnaire, there was no significant difference in the illusion rating between the congruent (M = 3.182) and incongruent (M = 3.000) stimulation trials, indicating that the subjective illusion of self-touch is equally compelling whether stimulation is congruent or incongruent. Schütz-Bosbach and colleagues (2009) demonstrated that the visual rubber hand illusion was resistant to violations of tactile expectation. In the incongruent trial of their study, the participant viewed a prosthetic hand being touched with either a soft or rough fabric while she received incongruent tactile stimulation on her hidden hand. We build on the results of Schütz-Bosbach and colleagues by demonstrating that the non-visual rubber hand illusion is also resistant to violations of tactile expectation. Participants experience the illusion of self-touch even under conditions of incongruent stimulation: the participant administers stimulation to the prosthetic hand using the index finger of her right hand while stimulation is administered to her left hand using a paintbrush. In the only prior study using the non-visual paradigm (Ehrsson et al., 2005), the incongruent stimulation trial took a different format. Specifically, the Examiner guided the participant to tap the bristles of a small dish brush while the Examiner synchronously stimulated the participant’s other hand. The illusion of self-touch was not typically elicited in this condition. There are some important differences between our Experiment 2 and the study by Ehrsson and colleagues (2005). In Experiment 2, the participant always administered stimulation to a prosthetic hand using her index finger but we manipulated the stimulus – index finger or paintbrush – the Examiner was using to touch the participant’s other hand. In contrast, in the study by Ehrsson and colleagues (2005), the researchers manipulated the object to which the participant administered stimulation – prosthetic hand or dish brush, but the Examiner always used an index finger to touch the participant’s other hand. 2 Earlier pilot experiments demonstrated that the illusion of self-touch could also be elicited when the participant administered stimulation with a paintbrush while her own hand was touched with a stick (and vice versa). However, in these conditions, the participant may have simply attended to administering touch with, and receiving touch from, an instrument; thus failing to register the stimulus incongruence. In the experiment proper, we used the finger-paintbrush incongruence condition, and so avoided this possible confound.
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
517
In short, we introduced incongruent stimulation by varying the instrument of stimulation while Ehrsson and colleagues (2005) introduced it by varying the object of stimulation. In our incongruent stimulation condition, the participant’s tactile expectations were violated but there was no conflict with her higher-order body representation. In this condition, participants experienced the illusion of self-touch. We suggest that, in the study by Ehrsson and colleagues (2005), participants did not experience the illusion of self-touch in the incongruent stimulation condition because they were unable to incorporate a dish brush into their higher-order body representation, whereas the prosthetic hand was more easily incorporated. One might question how the participant derives information that allows the prosthetic hand to be assimilated into her body image when vision of the hand is precluded. We believe that this information can be derived through stimulation of the prosthetic hand itself. The prosthetic hand used for the current experiments was exceptionally lifelike in shape and detail. When the participant administered stimulation, she was guided between the metacarpal phalangeal joint (the knuckle where the index finger meets the hand) and the proximal interphalangeal joint (the first knuckle of the index finger). The matching shape and structure of the stimulated object would have been sufficient to indicate its correspondence with the participant’s own stimulated hand. In Experiment 3, we introduced vision to assess whether visual feedback regarding stimulus incongruence would break the illusion of self-touch. There are two key differences in experimental design between the traditional visual rubber hand paradigm and the paradigm used in Experiment 3. First, in the traditional visual rubber hand paradigm, the participant is not involved in the administration of touch so there can be no illusion of self-touch, whereas in Experiment 3, the Examiner guided the participant to administer stimulation to the prosthetic hand. Second, in the traditional visual rubber hand paradigm, the participant is permitted vision of the prosthetic hand only, whereas in Experiment 3, the participant was permitted vision of either her right hand administering stimulation to the prosthetic hand or her left hand while it was being touched by the Examiner. Experiment 3 addressed two important questions: Can the illusion of self-touch be evoked when the participant has visual feedback? And, if the illusion can be evoked, will it persist when visual feedback indicates incongruent stimulation? A failure to experience the self-touch illusion in the visual condition may occur either because the illusion cannot be evoked with vision or because the participant does not experience the self-touch illusion irrespective of the visual manipulation. This second interpretation was discounted by testing participants from Experiment 1 who had been shown to experience the illusion in the non-visual version of the paradigm. In the congruent stimulation condition, participants did experience the illusion of self-touch when vision was permitted. Five participants (42%) indicated strong or very strong agreement with the statement ‘It felt as if I were touching my own hand’ even when they had visual feedback indicating that this was not the case. Note though that the illusion of self-touch was diminished by vision when the participant was looking at her right hand touching the prosthetic hand. How do we explain the finding that the illusion of self-touch was elicited whether the participant looked at her right hand touching the prosthetic hand or at her left hand being touched by the Examiner? In the view-right condition, the participant observed touch on the prosthetic hand which corresponded to the touch on her (hidden) left hand. The viewing conditions were thus matched to the traditional visual rubber hand paradigm, in which the Examiner administers stimulation to the prosthetic hand, but now the Examiner guided the participant to administer the observed stimulation to the prosthetic hand. When the participant views touch on the prosthetic hand that corresponds to touch on her own hand, she may experience visual capture of touch – the illusion that she is experiencing tactile sensations in the location of the viewed prosthetic hand (Botvinick & Cohen, 1998). The novel finding is that participants can also experience a visual rubber hand illusion when they are not looking at a prosthetic hand. In the view-left condition, the participant observed touch being administered to her own left hand by the Examiner. The action of the Examiner’s administering hand corresponded to the action of her (hidden) right hand which was administering touch to the prosthetic hand. We propose that, to experience the illusion that her hands are in contact, the participant may experience visual capture of action – the illusion that the action of her administering hand is in the location of the Examiner’s administering hand. Both types of displacement (touch and action) may lead to the visual illusion of self-touch. In the first case, tactile sensations are displaced to the location of the prosthetic hand and thus the location of the participant’s administering hand. In the second case, action is displaced to the location of the Examiner’s administering hand and thus the location of the participant’s touched hand. Note that the concept of visual capture of action is not new. Nielsen (1963) conducted an elegant experiment in which the participant completed a simple line-drawing task. Unbeknownst to the participant, a mirror was inserted into the experimental set-up so that the participant was viewing another person’s hand drawing the lines, rather than her own hand. Participants experienced the so-called alien hand as their own, making compensatory movements when the alien hand performed in an unpredictable manner. For example, when the task was to draw a straight line but the viewed hand veered rightward, most participants compensated for this error with a leftward adjustment of their own hand. Nielsen concluded that the ‘‘alien ‘visual hand’ dominate[d] the subject’s ‘kinesthetical/tactile hand’” (p. 230). He noted that most participants did not initially realise that they were viewing someone else’s hand. In considering Nielsen’s findings, we suggest that this pioneering study essentially evoked what would today be regarded as a rubber hand illusion, whereby participants took ownership of the actions of a viewed hand. (For a recent study demonstrating the illusion of ownership elicited using a moving rubber hand, see: Dummer, Picot-Annand, Neal, & Moore, 2009.) Having established that the illusion of self-touch can be elicited under congruent stimulation conditions which permit visual feedback, we next assessed whether the illusion was diminished under incongruent stimulation conditions which permit visual feedback. Specifically, we compared (a) the congruent trial in which the participant administered stimulation
518
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
with, and received stimulation from, an index finger and (b) the incongruent trial in which the participant administered stimulation with an index finger, and received stimulation from a paintbrush. There was no difference in illusion ratings between the congruent (M = 2.729) and incongruent (M = 2.583) stimulation trials and three participants (25%) indicated strong or very strong agreement with the statement ‘It felt as if I were touching my own hand’ even when they had visual feedback indicating incongruent stimulation. Experiment 3 demonstrates that when vision is permitted in the ‘non-visual’ rubber hand paradigm (in which the participant is involved in administering stimulation), participants experience an equally compelling illusion of self-touch for congruent and incongruent stimulation. A further question that we asked in Experiment 3 was whether participants would experience the illusion that they were observing their own right hand touching their own left hand, and whether this illusion would be affected by incongruent stimulation. The sense that one is observing one’s two hands in contact may be taken to suggest that the participant is experiencing illusory ownership over the ‘alien’ hand (i.e., the Examiner’s hand in the view-left condition or the prosthetic hand in the view-right condition). The illusion that the participant was observing her right hand touching her left hand was diminished by stimulus incongruence. Participants indicated higher levels of agreement with the statement ‘It seemed as if I were observing my right hand stroking my left hand’ in the congruent stimulation condition (M = 2.333) than in the incongruent stimulation condition (M = 1.479). How do we account for the diminished sense of ownership in the incongruent condition? When the participant looked to her left hand, the visual image was of a paintbrush touching her left index finger. This visual image was inconsistent with the tactile perceptions of the right hand, not only because of the incongruent stimulus but also because there was no guiding Examiner’s hand.3 Note that this latter point was also true of the congruent stimulation condition. When the participant looked to her right hand, the visual image was of her right index finger being guided to touch the prosthetic hand. This visual image was inconsistent with the tactile perceptions of the left hand. The results suggest that, although visual feedback of stimulus incongruence does not affect the felt illusion of self-touch, it does affect the illusion that the observed ‘alien’ hand is one’s own. 6. Conclusion In the current study, we investigated the role of tactile expectations in the self-touch illusion, without confounding tactile expectation and higher-order body representation. This was achieved by manipulating tactile properties of the administering stimulus (as in Schütz-Bosbach et al., 2009) rather than tactile properties of the object to which the participant administered stimulation. Experiment 1 established that the self-touch illusion was elicited when the participant administered stimulation to a prosthetic hand using a paintbrush or a stick while the Examiner administered stimulation to the participant’s other hand using a congruent stimulus. Previously, the self-touch illusion has only been evoked using stimulation administered with an index finger. In Experiments 2 and 3 we manipulated tactile expectations: the participant administered stimulation to the prosthetic hand using her index finger while the participant’s own hand was touched with either (a) a congruent stimulus – a finger or (b) an incongruent stimulus – a paintbrush. The self-touch illusion was not diminished by incongruent stimulation, and this was true whether the procedure was conducted with vision precluded or permitted. Armel and Ramachandran (2003) suggest that the traditional visual rubber hand illusion may be explained using a Bayesian model of perceptual learning, in which ‘‘two perceptions from different modalities are ‘bound’ when they co-occur with a high probability” (p. 1505). In the traditional visual rubber hand paradigm, the relevant modalities are vision (the observed touching of the prosthetic hand) and touch (the felt sensations in the participant’s hand). Thus, ‘‘the seen and felt touch were bound because of their temporal synchrony” (ibid.). A Bayesian model may likewise explain the illusion of self-touch, whereby the proprioceptive cues from the administering hand are bound with the tactile sensations of the receptive hand. Participants experience the illusion of self-touch when they administer touch to a prosthetic hand while receiving synchronous touch from the Examiner (see Ramachandran & Hirstein, 1997, 1998). According to Bayesian logic, there is a very low probability that the actions of the administering hand and the sensations on the receptive hand could correspond so precisely by chance; thus participants experience the illusion that the administering hand is touching the receptive hand. Self-as-active and self-as-receptive are experienced as participants in a single event of self-touch. Using incongruent stimuli (and indeed introducing vision) does not change the temporal correspondence between the actions of the administering hand and the sensations of the receptive hand, and the improbability that this correspondence occurred by chance. Armel and Ramachandran (2003) note that the brain takes advantage of statistical correlations, ‘‘even when they do not ‘make sense’ from the cognitive point of view” (p. 1505). In participants who experience the illusion of self-touch, information from multiple sensory sources is integrated into a single event file. Under conditions of incongruent stimulation, the dissonance between proprioceptive information from the administering right hand (I am administering touch with my finger) and tactile information from the receptive left hand (I am being touched with a brush) is apparent in participants’ descriptions of their experience. They may agree with the statement ‘‘My left index finger felt as if it were touched by a finger and a brush simultaneously” or with the statement ‘‘My right index finger felt like a brush”.
3
We thank an anonymous reviewer for this observation.
R.C. White et al. / Consciousness and Cognition 19 (2010) 505–519
519
Acknowledgments The authors would like to thank Mr Graham Thew for invaluable discussions, and two anonymous reviewers for helpful comments. References Armel, K. C., & Ramachandran, V. S. (2003). Projecting sensations to external objects: Evidence from skin conductance response. Proceedings of the Royal Society of London B, 21, 1499–1506. Botvinick, M., & Cohen, J. (1998). Rubber hands ‘feel’ touch that eyes see. Nature, 391, 756. Costantini, M., & Haggard, P. (2007). The rubber hand illusion: Sensitivity and reference frame for body ownership. Consciousness and Cognition, 16, 229–240. Dummer, T., Picot-Annand, A., Neal, T., & Moore, C. (2009). Movement and the rubber hand illusion. Perception, 38, 271–280. Durgin, F. H., Evans, L., Dunphy, N., Klostermann, S., & Simmons, K. (2007). Rubber hands feel the touch of light. Psychological Science, 18, 152–157. Ehrsson, H. H., Holmes, N. P., & Passingham, R. E. (2005). Touching a rubber hand: Feeling of body ownership is associated with activity in multisensory brain areas. The Journal of Neuroscience, 25, 10564–10573. Ehrsson, H. H., Spence, C., & Passingham, R. E. (2004). That’s my hand! Activity in premotor cortex reflects feeling of ownership of a limb. Science, 305, 875–877. Haans, A., IJsselsteijn, W. A., & de Kort, Y. A. W. (2008). The effect of similarities in skin texture and hand shape on perceived ownership of a fake limb. Body Image, 5, 389–394. Hägni, K., Eng, K., Hepp-Reymond, M.-C., Holper, L., Keisker, B., Siekierka, E., et al (2008). Observing virtual arms that you imagine are yours increases the galvanic skin response to an unexpected threat. Public Library of Science ONE, 3, e3082. Holmes, N. P., Crozier, G., & Spence, C. (2004). When mirrors lie: ‘Visual capture’ of arm position impairs reaching performance. Cognitive, Affective, and Behavioral Neuroscience, 4, 193–200. James, W. (1890). The principles of psychology. Cambridge, MA: Harvard University Press. Kinsbourne, M. (1995). Awareness of one’s own body: An attentional theory of its nature, development, and brain basis. In J. L. Bermúdez, A. Marcel, & N. Eilan (Eds.), The body and the self (pp. 205–224). England: MIT Press. Lloyd, D. M. (2007). Spatial limits on referred touch to an alien limb may reflect boundaries of visuo-tactile peripersonal space surrounding the hand. Brain and Cognition, 64, 104–109. Longo, M. R., Schüür, F., Kammers, M. P. M., Tsakiris, M., & Haggard, P. (2009). Self awareness and the body image. Acta Psychologica, 132, 166–172. Miles, E. (2008). What can the rubber hand illusion tell us about body representation in people with unexplained symptoms. In Symposium conducted at the Body Representation Workshop, University of Nottingham, UK. Moseley, G. L., Olthof, N., Venema, A., Don, S., Wijers, M., Gallace, A., et al (2008). Psychologically induced cooling of a specific body part caused by the illusory ownership of an artificial counterpart. Proceedings of the National Academy of Sciences, 105, 13169–13173. Mussap, A. J., & Salton, N. (2006). A ‘rubber-hand’ illusion reveals a relationship between perceptual body image and unhealthy body change. Journal of Health Psychology, 11, 627–639. Nielsen, T. I. (1963). Volition: A new experimental approach. Scandinavian Journal of Psychology, 4, 225–230. Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory. Neuropsychologia, 9, 97–113. Pavani, F., Spence, C., & Driver, J. (2000). Visual capture of touch: Out-of-body experiences with rubber gloves. Psychological Science, 11, 353–359. Pavani, F., & Zampini, M. (2007). The role of hand size in the fake-hand illusion paradigm. Perception, 36, 1547–1554. Ramachandran, V. S., & Hirstein, W. (1997). The three laws of qualia: What neurology tells us about the biological functions of consciousness, qualia and the self. Journal of Consciousness Studies, 4, 429–458. Ramachandran, V. S., & Hirstein, W. (1998). The perception of phantom limbs: The D. O. Hebb lecture. Brain, 121, 1603–1630. Schaefer, M., Flor, H., Heinze, H.-J., & Rotte, M. (2007). Morphing the body: Illusory feeling of an elongated arm affects somatosensory homunculus. NeuroImage, 36, 700–705. Schütz-Bosbach, S., Tausche, P., & Weiss, C. (2009). Roughness perception during the rubber hand illusion. Brain and Cognition, 70, 136–144. Shimada, S., Fukuda, K., & Hiraki, K. (2009). Rubber hand illusion under delayed visual feedback. Public Library of Science ONE, 4, e6185. Stevens, J. (1996). Applied multivariate statistics for the social sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum. Tsakiris, M., & Haggard, P. (2005). The rubber hand illusion revisited: Visuotactile integration and self-attribution. Journal of Experimental Psychology: Human Perception and Performance, 31, 80–91. Tsakiris, M., Hesse, M. D., Boy, C., Haggard, P., & Fink, G. R. (2007). Neural signatures of body ownership: A sensory network for bodily self-consciousness. Cerebral Cortex, 17, 2235–2244. Tsakiris, M., Prabhu, G., & Haggard, P. (2006). Having a body versus moving your body: How agency structures body-ownership. Consciousness and Cognition, 15, 423–432.
Consciousness and Cognition 19 (2010) 520–533
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Edges, colour and awareness in blindsight Iona Alexander *, Alan Cowey University of Oxford, Department of Experimental Psychology, South Parks Road, Oxford OX1 3UD, UK
a r t i c l e
i n f o
Article history: Received 25 September 2009 Available online 18 February 2010 Keywords: Blindsight Colour contrast Luminance contrast Narrow-band colours
a b s t r a c t It remains unclear what is being processed in blindsight in response to faces, colours, shapes, and patterns. This was investigated in two hemianopes with chromatic and achromatic stimuli with sharp or shallow luminance or chromatic contrast boundaries or temporal onsets. Performance was excellent only when stimuli had sharp spatial boundaries. When discrimination between isoluminant coloured Gaussians was good it declined to chance levels if stimulus onset was slow. The ability to discriminate between instantaneously presented colours in the hemianopic field depended on their luminance, indicating that wavelength discrimination totally independent of other stimulus qualities is absent. When presented with narrow-band colours the hemianopes detected a stimulus maximally effective for S-cones but invisible to M- and L-cones, indicating that blindsight is mediated not just by the mid-brain, which receives no S-cone input, or that the rods contribute to blindsight. The results show that only simple stimulus features are processed in blindsight. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Blindsight is the ability, possessed by many patients following destruction of the striate cortex (V1), to detect, localise and even discriminate among visual stimuli in their clinically blind field by ‘guessing’, despite not consciously experiencing a visual percept. Occasionally the patients say that they are aware that something happened but that it was not a visual percept, named type 2 blindsight by Weiskrantz (1998). In this paper we do not formally consider this distinction. Despite three decades of research it is still unclear which stimulus properties sustain this ability and precisely where the processing takes place. For example, the ability of hemianopic monkeys and patients to detect stimuli whose space-averaged luminance is the same as the background, i.e. isoluminant, has been repeatedly demonstrated (for reviews see Cowey (2004), Stoerig (2006) and Stoerig and Cowey (1997). However, the stimuli had either a spatially sharp luminance or chromatic boundary with respect to their background and it may be the ability to detect sharp luminance or chromatic borders that survives in blindsight rather than perception of other qualities of the entire stimulus itself, such as surface colour or texture or brightness or shape. In this paper we describe five experiments, one on luminance and four on colour, in order to elucidate the properties of the pathways that subserve the ability of hemianopic subjects to detect and discriminate such unseen stimuli in their blind field. Visual information is transmitted from the retina to the brain via nine direct projections (see Cowey (2010), Cowey & Stoerig (1991), Stoerig (2006), and Stoerig and Cowey (1997) for reviews) and visual information can reach and be processed at further stages in the brain via the interlaminar layers of the dLGN, the superior colliculus, the pre-tectum and the pulvinar. These connections have been explored extensively (Stoerig & Cowey, 1997), but their role in blindsight, and how some stimuli sustain blindsight more readily than others is still contested. Traditionally, residual visual functions in monkeys were
* Corresponding author. Fax: +44 1865 310447. E-mail address:
[email protected] (I. Alexander). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.01.008
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
521
attributed to mediation by the scotopic rod system (Klüver, 1942, 1949; Leporé, Cardu, Rasmussen, & Malmo, 1975; Malmo, 1966), but Humphrey (1974) reported that destriate monkey Helen could navigate her complex outdoor environment in daylight, which surely requires more than rod vision! She certainly did not behave like human rod monochromats, who are severely visually disabled in bright light. However, it must be acknowledged that their vision is not identical to normal rod vision. As well, Schilder, Pasik, and Pasik (1972) and Keating (1979) reported that monkeys with total removal of V1 could discriminate between different wavelengths, as can human subjects with blindsight in their field defects (Stoerig, 1985), even when luminance was varied in order to make it irrelevant. Cowey and Stoerig (1999) and Stoerig and Cowey (1989) showed that rod and cone mechanisms operate in both the normal and the hemianopic fields of patients and monkeys, and that the blind field even shows a Purkinje shift, indicative of both rod and cone function, results also shown by Brent, Kennard, and Ruddock (1994) in patient GY. The retino-collicular pathway, rather than the koniocellular or pulvinar LGN pathways, has been implicated in colour processing in blindsight. However, electrophysiological evidence (de Monasterio, 1978; Marrocco & Li, 1977; Schiller & Malpeli, 1977) indicates that there is no direct retinal input from the S-cones to the colliculus, implying that colour discrimination should be impaired when it requires S-cone, blue/yellow, processing. Indeed, Sumner, Adamjee, and Mollon (2002), using the same reasoning, demonstrated that stimuli visible only to the S-cones failed to produce the saccadic distraction widely attributed to the superior colliculus. But it should be acknowledged that more recent evidence (Hall & Colby, 2009) does point to a pathway from S-cones to the superior colliculus. Ignoring the latter for the moment, if chromatic information in blindsight is processed via the superior collicular, rather than a pulvinar or koniocellular LGN pathway, blindsight should be insensitive to stimuli that isolate the S-cone mechanism. This prediction is born out in patients who have blindsight following complete hemispherectomy (Leh, Mullen, & Ptito, 2006; Ptito, 2007) but might not be true for patients in whom the blindsight followed damage only or chiefly to striate cortex. We therefore tested two such subjects with narrow-band coloured stimuli, one of which would be barely detectable by M- and L-cones. The experiments were designed to investigate the role of simple stimulus features such as colour, luminance, edges and temporal onsets in facilitating the ability of blindsight patients to detect stimuli in their blind field and accordingly to shed light on what is being processed in blindsight. 2. Subjects Two hemianopes (GY and MS) were studied. Both took part in the target localisation, and narrow-band stimuli studies but only GY in the colour discrimination experiments because MS’s V1 lesion additionally and bilaterally destroyed the ventral temporal region concerned with colour vision and previous publications showed that he was unable to discriminate between colours even in his seeing hemianopic field (reviewed by Heywood and Cowey (2003)). In all four experiments the stimuli were displayed on a computer screen and eye fixation was monitored on a large screen, visible to one of the two experimenters by means of an infra-red camera at the side of the display throughout all trials. Any trials where fixation was not maintained during stimulus presentation were eliminated but this rarely occurred. Hemianope GY has been reported in detail elsewhere (e.g. Barbur, Ruddock, & Waterfield, 1980; Barbur, Watson, Frackowiak, & Zeki, 1993; Azzopardi & Cowey, 2001; Brent et al., 1994; Bridge, Thomas, Jbabdi, & Cowey, 2008; Cowey & Walsh, 2000). In brief, he suffered a unilateral lesion in his left medial occipital cortex caused by a traffic accident when he was 8 years old. In his left hemisphere the occipital pole is the only part of his striate cortex to survive, accounting for his macular sparing of about 3°. GY has been extensively examined for his residual visual function in his hemianopic field and he retains the ability to detect, localise and discriminate stimuli there. But even when he reports being aware of a stimulus he denies having a visual percept, referring instead to ‘a feeling that something happened’. Subject MS has also been reported in detail elsewhere (reviewed by Heywood and Cowey (2003)). In brief, MS contracted herpes encephalitis in 1971 which destroyed most of the ventral temporal cortex of both hemispheres and in addition the calcarine cortex on the right, leaving him with a complete left homonymous hemianopia. In his intact hemifield he is agnosic for faces and objects and has total achromatopsia, which has been studied extensively. Surprisingly he has not hitherto been tested for any residual visual capacity in his blind field, apart from his pupillary response, which is intact (Cowey, Alexander, Heywood, & Kentridge, 2008). He was aged 61 at the time of the current investigation. Control subjects were not needed as both experimenters could perform all tasks faultlessly. 3. Experiment 1: localisation task Stimuli were generated using a Cambridge Research Systems VSG II/5, programmed with a Dan computer and visual basic software. Stimuli were displayed on an EIZO 19inch monitor, calibrated using OPTICAL (Cambridge Research Systems), with a viewing distance of 57 cm. The five stimuli were: a plain white 8 8° white square, a vertical square-wave grating (contrast 0.5) of the same size as the white square and with a mean luminance the same as the background, a Gaussian of the same peak luminance as the square, a Gaussian of the same mean luminance as the square but of higher peak luminance, and a Gabor of the same mean luminance and contrast as the grating (see Fig. 1). Background screen luminance was 10 cd/m2, the same as the mean luminance of the gratings and the second Gaussian and the Gabor. The square was 25 cd/m2 and the peak luminance of the two Gaussians was 25 or 50 cd/m2. Each block of one stimulus type contained 100 trials and GY completed two blocks of each stimulus condition. The parameters for the Gaussian, Gabor and gratings were determined using VSG
522
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
Fig. 1. Examples of the stimuli in the localisation paradigm. Top left: plain square; bottom left: Gaussian; top right: square-wave grating; bottom right: Gabor patch. Subjects completed 100 trials with each stimulus type, and each stimulus appeared in any one of the four locations at random. For subject MS the stimuli were 1.7 times larger in linear extent but centred on the same position. The central spot marks the start light.
software and the spatial frequency of the grating and Gabor was 0.75 cpd and the radial size of each standard deviation of the Gaussian patch was the height of the screen in pixels divided by 20, and for the Gaussian it was set to 40. With GY the stimuli were presented in either the upper or lower quadrants, at random, in both the left and right hemifields. Because of time constraints the task was simplified for MS by making the stimuli larger, 13 13°, and presenting them only in the blind hemifield after establishing that he could perform the task perfectly when they were in the seeing hemifield. 3.1. Methods On each trial the subject fixated a white start light that appeared at the centre of the VDU. When the subject pressed the space bar on the keyboard, the start light disappeared and was instantaneously followed by a 200 ms stimulus in one of the four quadrants of the VDU, either in the good field (upper or lower) or in the blind field (upper or lower). Stimuli were equiprobable in both visual fields for GY, but were predominantly in the blind field for MS. The subject started each trial and was instructed to respond as quickly and accurately as possible using the keyboard responses; upper left ‘q’, lower left ‘z’, upper right ‘p’, lower right ‘m’. The four keys were prominently marked by blobs of plasticene so that the subject could rest four fingers on them throughout testing and did not have look at them. Examples of the displays are shown in Fig. 1. MS found the keyboard response too confusing and was only comfortable with a verbal response. Accordingly, the experimenter entered MS’s verbal responses on the keyboard. Manual reaction time data were therefore not collected for MS. After each block of trials the two hemianopes were asked whether they had experienced any kind of awareness of the stimuli but this was not done after every trial, given the large number of trials and constraints on time. 3.2. Results When performance with the Gabor stimulus, and its control stimulus (square-wave grating) were compared there was, as predicted, a significant difference in performance for both GY (v2 = 24.355; df = 1, p < .001, 1-tailed) and MS (v2 = 3.945; df = 1, p < .05, 1-tailed). A binomial analysis demonstrated that MS’s performance with the Gabor was no better than expected by chance (p > .05). A one-way ANOVA revealed a large and significant difference in GY’s reaction times (F(1, 163) = 40.180, df = 1, p < .001). GY’s mean reaction time for the Gabor was 818.01 ms and for the grating 545.93 ms (Fig. 2 right panel), i.e. GY was much slower to respond to the stimulus which lacked sharp luminance contours. The reaction time in the good field for the Gabor was 681.25 ms and for the grating was 549.74 ms. There was a significant difference between the percentage correct for the square and the first Gaussian, whose peak luminance was matched to that of the plain square, for GY (v2 = 10.940; df = 1, p < .001, 1-tailed) and for MS (v2 = 32.175; df = 1, p < .001, 1-tailed). Binomial analysis demonstrated that MS’s performance with the Gaussian was no better than expected by chance (p > .05). There was a significant difference between percentage correct for the second Gaussian, mean luminance matched to that of its control the square 20 compared to the plain square for GY (v2 = 9.872; df = 1, p < .001, 1-tailed) but not for MS (v2 = 2.240; df = 1, p > .05, 1-tailed). There was a significant difference between GY’s reaction times for the plain square (578.4 ms) compared to the Gaussian (674.76 ms) (F(1, 170) = 2.845, df = 1, p > .05) and between the plain square and the Gaussian20 (672.60 ms) (F(1, 171) = 17.664, df = 1, p < .001) (see Fig. 2 left panel). The reactions times in the good field for the plain square was 601.80, for the Gaussian 651.51 and for the Gaussian20 597.42.
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
523
Fig. 2. Right panels: results of the localisation task using a square-wave grating stimulus or a Gabor patch. Performance declined for GY and was reduced to chance level for MS when the stimulus lacked sharp contours. Mean reaction time was also significantly longer for GY in the Gabor condition. Left panels: results with a flashed plain white square or a Gaussian white patch of the same peak luminance or the same mean luminance. Performance was impaired for GY, but his reaction time was longer in responding to the Gaussian. MS localised the flashed square well but was totally unable to localise the Gaussian until it’s peak luminance was raised to 20 cd/m2.
4. Experiment 2: red/green colour discrimination using Gaussian patches Stimuli, generated using the VSG II/5 system as described in experiment 1, were 10° in size and presented in the blind hemifield straddling the horizontal meridian. The stimuli were coloured Gaussian patches of either red or green, presented on a grey background. Each stimulus lasted for 500 ms followed by a 50 ms interval and then another Gaussian patch of either the same colour or the complementary (green or red) colour for 500 ms (Fig. 3). Subject GY, had to indicate whether or not the Gaussian had changed colour. There were five conditions, where luminance was varied either for the background or for the red or the green Gaussian (see Table 1). Note that GY did not have to name the colour, merely to say whether it had changed. Next, in order to assess whether GY could determine whether a Gaussian patch was green (9 cd/m2) or red (9 cd/m2) on a grey background of 9 cd/m2, we used a similar procedure but this time only one coloured Gaussian was presented, for 1000 ms, and GY verbally reported ‘red’ or ‘green’. Following this, for 100 trials we presented a single red or green Gaussian of unlimited duration while the screen was covered by a white card before slowly moving the card upwards over a period of about 2 s to uncover the complete Gaussian, which then had to be categorised as red or green. Only then was the stimulus turned off. Finally, for 50 trials, the screen was no longer obscured by the card. Instead GY had to close his eyes, and then open them after the stimulus appeared on the screen as indicated by the experimenter. The purpose of both procedures was to minimise any transients present when the entire stimulus is presented in a single frame and to eliminate all transients caused by stimulus offset. 4.1. Results GY could discriminate whether or not the colour had changed at every ratio of red/green luminance and at two different background luminances when the stimuli were briefly flashed (Fig. 4). Although he performed least well when the two stimuli were photometrically isoluminant against a background of much higher luminance his score of 87% correct was only just significantly different from the condition where the background was the same luminance as the stimuli (Fig. 4, top left, v2 = 3.911; df = 1, p < .05). All other comparisons were insignificant and he scored close to 100% correct (Fig. 4 top right and bottom left). Despite denying any visual percept he described the red stimulus as ‘‘having a bigger effect”, which is why the luminance of the green stimulus was increased to 20 cd/m2 in an attempt to make it even more salient so that it matched the red in salience. But he still described the red as being easier to detect. Nor was there any significant change in his excellent performance when the red luminance was reduced from 9 cd/m2 to 6 cd/m2 and the green was 20 cd/m2. These results indicated that GY could tell the difference between a flashed red and green even when they had a Gaussian
524
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
Fig. 3. Examples of the Gaussian blob sequences used in experiment 2. The subject indicated whether or not the Gaussian blob had changed colour in a 2interval task.
Table 1 Stimulus luminance of the background and the red and green stimuli in experiment 2. Condition
1 2 3 4 5
Luminance (cd/m2) Background
Red
Green
18 9 9 9 9
9 9 9 9 6
9 9 12 20 20
profile. However, it became evident that he was detecting differences in the salient stimulus onset or offset, or both. When the stimuli and the background were all isoluminant and now only one colour was presented on each trial and it remained there until GY responded by naming it red or green he scored 100% correct and explained that he could do it because the red produced a stronger feeling whereas the green did not. But when the stimulus was presented behind the white card before being slowly uncovered he scored only 54/100. The difference between the scores for the two conditions was highly significant (v2 = 17.160; df = 1, p < .001, see Fig. 4, bottom right). When asked about this he said that when the stimulus was
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
525
Fig. 4. Percentage correct for the red/green colour discrimination task. GY could detect a sudden change from green to red or visa versa but this ability was abolished when the stimulus onset was gradual or completely obscured. Top left: background luminance varies, bottom left: green luminance varies, top right: red luminance changes, bottom right: onset varies.
flashed he could detect an event and that the events were different for red and green. But he could detect nothing when the identical coloured stimuli were slowly uncovered, no matter what the colour on any trial (Fig. 4, bottom right). Since it could be argued that the slow removal of the card might have ‘masked’ the coloured stimulus, the procedure was repeated by presenting the stimulus immediately after he closed his eyes, which he then re-opened on command. He was still unable to discriminate red from green (22/50) because, he said, ‘‘neither of them produced any feeling” (Fig. 4, bottom right). 5. Experiment 3: edges and temporal onsets In order to assess the role of stimulus transients and edges on stimulus detection we used stimuli with temporal envelopes designed to minimise any transients present when the entire stimulus is presented in a single frame, and used stimuli with and without Gaussian envelopes to evaluate edge detection mechanisms. A forced choice detection paradigm was used and on each trial the subject indicated whether or not a stimulus was presented. There were two types of stimuli: a circle with sharp edges or a Gaussian i.e. without sharp edges. The stimulus was either red (9 cd/m2, x = 0.596, y = 0.346) green (9 cd/m2, x = 0.291, y = 0.556) on a grey background (9 cd/m2, x = 0.300, y = 0.316) or blue (5.5 cd/m2, x = 0.151, y = 0.075) on a grey background (5.5 cd/m2, x = 0.300, y = 0.311), and the onset was either sudden or the stimulus reached its peak slowly, over 1 s. When there was no offset, the stimulus remained on the screen until a response was made. When there was an offset the stimulus remained at its peak for 200 ms and then disappeared suddenly. Four conditions were used: (1) a stimulus with sharp edges and a sudden onset, but no offset; (2) sharp edges with slow onset of 1 s, but no offset; (3) sharp edges and slow onset of 1 s but sudden offset and (4) a Gaussian with a sudden onset and no offset. The stimuli were 8° in diameter and presented in the blind field with their nearest ‘edge’ 9° lateral to the fixation spot. There were 50 trials per condition, per colour. 5.1. Procedure The subject looked at the fixation spot and when the experimenter said ‘now’ the trial was initiated and a stimulus appeared in the blind field, in which case the correct response was ‘yes’, or there was no stimulus and he had to respond ‘no’. Responses were recorded manually and percentage correct calculated. The fixation mark remained on the screen throughout and eye movements were monitored. The subjects were told that there would be a stimulus on half of the trials at random and that responding ‘No’ on every trial was inappropriate; the procedure was therefore Yes or guess.
526
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
5.2. Results Fig. 5 shows that when the stimuli had sharp edges and a sudden onset GY was able to detect the red, green and blue stimuli (binomial tests, all p < .001). But this ability was abolished when the temporal onset was slow (1 s) and the stimulus was either green or red (all p > .05), but not when it was blue (p < .001). When the stimuli had a slow onset, but a sudden offset performance was re-established for the red (p < .001), but not for the green (p > .05). Thus GY can use the onset or the offset for stimulus detection. He could detect a red or blue (p < .001 in both cases) but not a green (p > .05) Gaussian. (Fig. 5, bottom). MS was able to do two of the 12 discriminations. He could accurately detect a green Gaussian stimulus (p < .05) and a blue stimulus (p < .001) with a sharp onset and no offset. Unlike GY, MS was unable to detect a stimulus with a slow onset and a sudden offset, even for blue stimuli. (Fig. 5 top). 6. Experiment 4: colour discrimination: red/green and blue/yellow This experiment was carried out only with GY. Visual stimuli were presented on a white background on 17-in colour monitor incorporating a touch screen (Phillips UP2799) at a viewing distance of 28 cm. The stimulus could appear in any
Fig. 5. Pecentage correct scores in detecting a red, green or blue stimulus with sharp or Gaussian spatial profile, and different temporal onsets and offsets. Only GY could perform well with several of the conditions but even he was unable to achieve better than chance performance with a green stimulus unless its contours were sharp and its onset sudden and he was equally poor with a red stimulus whose onset was slow. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
527
one of the four quadrants at random, as in experiment 1. The luminance and chromaticity of each stimulus was measured as before. On each trial a single stimulus was presented for 200 ms and was designated either positive or negative. A go/no-go procedure was used. If the stimulus was positive GY had to touch its remembered position as quickly as possible i.e. a go trial. If it was negative he had to refrain from responding for 3 s until the trial ended. i.e. a no-go trial. For each pair of colours the positive stimulus remained the same luminance throughout the experiment but the luminance of the negative stimuli was changed after each block of 100 trials in order to determine whether GY could discriminate between the two colours irrespective of their relative luminance. Two discriminations were used; red (positive, go trial) vs. green (negative, no-go trial) and blue (positive, go trial) vs. yellow (negative, no-go trial). 6.1. Procedure On each trial GY had to touch the start light at the centre of the screen. This initiated the trial and produced a 200 ms target in one of four quadrants; top left, bottom left, top right or bottom right. If it was the positive target GY had to respond by touching its remembered location; if it was negative, he had to refrain from responding. Incorrect responses were instantly signalled by turning the display black. Correct touches were signalled by filling the previous stimulus area with bright white light for 1 s. The inter-trial interval was 4 s, after which the start light re-appeared for the next trial. Trials were selfpaced and there were 100 trials with each condition. 6.2. Results In this experiment the display was the same as in experiment 2 but only one stimulus was presented, for 200 ms, in any one of the four quadrants on each trial and GY had to respond by touching its remembered position if it was positive and refrain from responding if it was negative. The stimuli were spatially of uniform colour and luminance with a sharp chromatic border with the 10 cd/m2 background. The luminance of the standard red stimulus was 2 cd/m2 in a further attempt to reduce its salience. When the luminance of the green stimulus was titrated down in relation to the red stimulus, GY remained able to detect a difference between red and green when green was 46 cd/m2, 27 cd/m2, 18 cd/m2 (p < .001), 5.5 cd/m2 or 2.3 cd/m2 (p < .01), but not when green was 13.5 cd/m2 or 7.5 cd/m2 (p > .05). In the latter conditions he said that he was still aware of the stimuli but that they produced the same ‘feeling’. When a similar procedure was used with blue positive, 5.5 cd/m2, and yellow negative and the luminance of the yellow was titrated, his excellent performance when yellow luminance was 5.5, 6.5. 10, or 28 cd/m2 (about 70–90% correct, p < .001 in all cases) was replaced with chance levels of responding (when it was 12.4, 14, 18.5 or 23 cd/m2 (p > .05). Again in the difficult pairings he reported that ‘‘the two colours gave him the same impression”. In other words he was responding to salience rather than only to hue, the salience varying with the luminance of the colours (Fig. 6). Of course, salience might be mediated by an interaction between the hue and luminance of the stimuli if both are being processed in blindsight.
7. Experiment 5: narrow-band colours There is evidence that the S-cones provide no direct input to the superior colliculus (e.g. Sumner et al., 2002). If blindsight is mediated chiefly by the superior colliculus, as some experiments indicate, blindsight should not be present with stimuli that can only be detected by S-cones. Experiment 5 investigated this. Unfortunately VDU phosphor emissions are broad-band and any ‘pure’ blue stimulus generated by exciting only the blue gun will stimulate not only S-cones but also L and M-cones, albeit to a lesser degree. In order to circumvent this, a tritanic confusion line (Smithson, Sumner, & Mollon, 2003) for any given subject can be calculated, thus ensuring that stimuli really do isolate the S-cones. Although this is straightforward in subjects with normal vision, it is very difficult to achieve in blindsighted subjects. One could assume that the two hemifields are similar with respect to their chromatic properties, in which case the tritanopic confusion line could be determined in the good field, then translated to the blind field. However, there is extensive transneuronal retrograde retinal degeneration of retinal ganglion cells following ablation of the striate cortex and there is therefore no assurance that the two hemifields are the same. Furthermore, as this retinal degeneration is selective for the colour-opponent P ganglion cells (Cowey, Stoerig, & Perry, 1989), it would not be surprising if the tritanopic confusion lines in blindsight and real sight differed substantially. Accordingly we avoided the problem of trying to match the line in the two hemifields and instead we assessed whether GY and MS could detect or discriminate a range of narrow-band stimuli in their blindfield. This was done by using a series of 5 5 cm interference filters (Schott, Glaswerke) which transmitted light of different peak wavelengths with a band-width of 10–12 nm at half height. The filters were calibrated by a spectrophotometer (Perkin–Elmer Lambda Series/PECSS) and their characteristics are shown in Fig. 7 (top). A Kodak carousel S-AV 2020 projector was mounted immediately behind the subject’s shoulder on his blind side, making it possible to project an intense beam of narrow-band light on a large white card 57 cm in front of the subject. The shutter in the projector could be programmed to deliver a stimulus for any period of time. The circular stimulus subtended 10° for GY and 20° for MS and its nearest edge in the blind field was 10° from the fixation point on the white card. Focal lighting was arranged so that the sighted hemifield was flooded with white light at a mean intensity of about 70 cd/m2, leaving the background of the blind hemifield in the region of the stimulus at 10 cd/m2 for GY and 2 cd/m2 for MS. Eye movements were monitored throughout.
528
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
Fig. 6. Results of experiment 4. Discrimination between red and green and between blue and yellow at different relative luminances of the two stimuli, which were presented singly. GY was unable to score at better than chance levels at certain luminance ratios, indicating that colour discrimination independent of other stimulus properties was absent.
7.1. Procedure On each trial, signalled verbally by the experimenter, a single stimulus was presented for 500 ms, either before or after a blank stimulus (i.e. no stimulus) for 500 ms, at random and with an interval of 500 ms between them. The subjects had to report whether the stimulus appeared in the first or second interval. Six different filters were used (Table 2) with the same filter throughout each block. In addition we carried out three further conditions on patient GY. First, GY had to guess whether the narrow-band stimulus was red or blue (18 cd/m2). Second, in order to minimise the transients present at stimulus onset we repeated this condition but GY closed his eyes prior to stimulus presentation, and opened them only after the stimulus had appeared (as signalled by the experimenter), and indicated whether the stimulus was red or blue within 5 s (before the stimulus was removed). Finally he was asked to repeat the previous condition, but on half the trials the stimulus was red as before and on half it was blank, rather than blue. GY indicated whether there was a stimulus or not. The purpose of this test was to minimise any transients present when the entire stimulus is presented in a single frame and to eliminate all transients caused by stimulus offset. 7.2. Results As shown in Fig. 8. GY scored 100% correct for 4 of the coloured stimuli and his poorest score was 87% with the deep blue stimulus that peaked at 426 nm. MS was less successful but still scored between 65% and 83% with all the stimuli other than deep red (Fig. 8). After each block of 100 trials with a given stimulus both subjects were asked what they had experienced, if anything. GY was confident that he could tell that ‘‘something happened in his blind field” with all the colours except deep blue 426 nm and green 514 nm, which he said hardly ever elicited any experience; yet he scored 87% with the former and 100% with the latter, which amounts to pure guesswork, i.e. blindsight. In contrast, MS struggled to describe how he decided on the correct interval and said he ‘‘thought he might know the right one”. He often took several seconds to make up his mind
529
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
Fig. 7. Top: transmission characteristics of the seven narrow-band stimuli used in experiment 5. The four symbols near the top of the curves indicate, from left to right, the absorption maxima for the S-cones, rods, M-cones and L-cones. The numbers indicate the filter reference number. Bottom: mean absorbance spectra for the four human photoreceptors (redrawn from Dartnall, Bowmaker, & Mollon, 1983); filled circles, rods; squares, S-cones; triangles, M-cones; plain circles, L-cones.
Table 2 Properties of narrow-band stimuli used in experiment 4. Filter
Peak wavelength (nm)
Peak transmittance
cd/m2
4 2 6 5 1 1 3
426 462 464 514 529 529 629
.51 .49 .58 .59 .50 .54 .35
2.1 12 18 12 12 18 18
and said that he ‘‘was just doing his best to guess”. No feedback about performance was given until the end of each block and GY thought he had scored no better than chance with the two colours that elicited no feeling. MS was pleased to find that he had scored so well. Unlike GY, MS did not score better than expected by random responding to the deep red stimulus, a point taken up in the discussion. Finally, and revealingly, GY scored 46/50 when asked to discriminate between a red and blue narrow-band stimulus, but only 24/50 when he was asked to close his eyes, and open them after the onset of the stimulus (indicated by the experimenter). Repeating this paradigm but with a red and blank, rather than red and blue stimulus, demonstrated that GY was unable to detect the presence of the stimulus (23/50) under these conditions. 8. Discussion The performance of both hemianopic subjects on a variety of visual discrimination tasks varied substantially, even within a particular discrimination paradigm, according to changes in the luminance, contrast, wavelength and onset–offset
530
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
Fig. 8. Performance on two-interval forced-choice task with narrow-band stimuli. Both GY and MS could discriminate short-wavelength stimuli in their blind field, but unlike GY, MS could not detect or discriminate red (629 nm).
characteristics of the stimuli. The five main tasks are discussed separately before a general discussion of the implications of the results for ideas about blindsight. 8.1. Experiment 1: target localisation GY’s reaction time in the localisation task was lengthened when stimuli lacked any sharp luminance borders, as with Gabors and Gaussians, but his performance in both conditions was significantly above chance; it was merely impaired rather than abolished under these conditions. This is contrary to MS’s performance, which was reduced to chance levels with both the Gabor and Gaussian stimuli compared to their sharp-edged controls. Not only does this highlight the importance of sharp boundaries, it shows that the latter stimuli could not have been detected by stray light because discernable light scatter would be similar for both stimulus pairs. As eliminating the sharp boundaries significantly impaired rather then abolished performance in GY, it is impossible to determine from this experiment whether other properties such as colour and texture may also contribute to an object’s salience. Although the sharp edge was eliminated it is possible that GY could not detect the Gaussian function from its onset to peak in its entirety, but rather from a point along that curve, thus resulting in a perceived ‘smaller’ stimulus. By varying the point of interpolation of the Gaussian or Gabor function it should be possible to investigate the existence of a null point, and the sensitivity of edge detection mechanisms in blindsight with varying stimulus edges, and determine whether it is possible to impair blindsight, perhaps to the point of abolishing it in this way. Whatever the outcome of such further experiments these results indicate that processing stimulus properties such as colour, surface or texture is intact, and that the ability of blindsight patients to detect isoluminant stimuli is not solely attributable to luminance boundaries. Unlike GY, MS described himself as ‘‘flummoxed” by the elimination of sharp borders in the stimuli, and he performed no better than chance. This is consistent with the extensive damage to his extra-striate ventral visual pathway, which is intact in GY, and his denial of being aware of any stimulus. 8.2. Experiment 2: coloured Gaussians GY could detect a colour change from red to green or vice versa when both stimuli had sharp temporal onsets and offsets, and he could correctly name a single colour as red or green. But the latter ability was abolished when the stimulus was slowly uncovered, regardless of whether the colour was red or green, and was similarly abolished when the stimulus was presented with his eyes closed and he had to identify the colour after opening his eyes while the stimulus remained in view for up to 5 s. This suggests that the onset and/or offset of the stimulus, as well as the sharpness of its boundary, is paramount. Thus, although Gabor and Gaussian patches of luminance stimuli impaired blindsight (experiment 1), they did not do so with the coloured stimuli used in this experiment. Although this was a discrimination task, rather than a localisation task, coloured stimuli appear to have their own salience, which further supports the idea that some chromatic processing mechanisms are intact in blindsight.
531
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
8.3. Experiment 3: edges and temporal onsets The results could hardly be clearer. MS only responded reliably to the blue stimulus with sharp spatial and temporal onset. GY performed more effectively, i.e. to blue stimuli whatever their spatial and temporal properties and to red except when it had a slow onset. As with luminance stimuli (Barbur, Harlow, & Weiskrantz, 1994) the subjects were responding to spatial and temporal properties of the coloured stimuli rather than to their chroma, as confirmed in experiment 4. The excellent performance with blue stimuli may indicate that it is mediated by rods, whose quantum catch at mesopic viewing conditions will be substantial. 8.4. Experiment 4: colour discriminations Green/red: GY could usually discriminate the difference between red and green. But when the red stimulus was 2 cd/m2 his performance was impaired when the green was 13.6 cd/m2 or 7.5 cd/m2 (p > .05) although not when the green was at the higher values of 46 cd/m2, 27 cd/m2 18.3 cd/m2or the lower values of 2.3 cd/m2 and 5.5 cd/m2 (p < .01). Thus the mechanism of red/green colour discrimination in GY must be different from that in normal vision in that for GY there is a luminance ratio at which the difference between green and red becomes undetectable. Further, at this point GY reported that the stimuli produced the same ‘feeling’, i.e. their salience became matched. Interestingly, when cone contrasts are calculated for each stimulus (i.e. (bgL stL)/(bgL + stL) for the L-cones and (bgS stS)/(bgS + stS) for the S-cones where bg = background and st = stimulus) the results can be predicted according the S-cone excitation, i.e. the stimuli GY was unable to tell apart (red 2 cd/m2 and green 13.6 cd/m2) have the smallest difference between the S-cone excitation induced by red or green (see Table 3). Whether rods might also be activated is discussed below. Blue/yellow: As with green and red, there was a luminance ratio at which discrimination of blue from yellow became impossible for GY in his hemianopic field. Although he could discriminate between blue and yellow when the yellow was 5.5, 6.5, 10, 23 and 28.5 cd/m2 (p < .01) he failed to do so when it was 12.5, 14 or 18.5 cd/m2 (p > .05). This suggests that blue – like red – is a particularly salient stimulus for him but that if yellow is made much more luminously intense it becomes as salient as the blue and accordingly indistinguishable from the blue. 8.5. Experiment 5: narrow-band stimuli Both GY and MS could discriminate and detect short wavelength blue stimuli in their blind field. Unlike GY, MS was unable to detect or discriminate red. This is contrary to what was expected if blindsight is entirely mediated by the superior colliculi, which lack any input from S-cones, and if the narrow-band blue which peaked at 427 nm, does not stimulate Lcones which peak at 564–580 nm (with a range of 500–700 nm) or M-cones which peak at and 534–545 nm (range 450– 630 nm) (see Fig. 7 bottom). Either there are other pathways, as argued by Stoerig and Cowey (1992) and Cowey and Stoerig (1999) who found that spectral sensitivity of hemianopic patients and macaque monkeys was not selectively impaired at short wavelengths or, alternatively, GY and MS were detecting the narrow-band stimuli via their rods, whose peak absorbance is 498 nm and which do contribute to the collicular input. All the testing was carried out at mesopic adaptation levels and rods must have contributed. All investigators of residual visual sensitivity following damage to V1 (from Klüver, 1942, onward) have found that low levels of ambient light assist, or are even necessary for, residual visual processing in the blind field. This might also explain why MS, whose lesion is very much larger and destroys almost the entire ventral temporal cortex, could not detect or discriminate the narrow-band red stimulus, which peaked at 630 nm, well outside the effective range of rods. Perhaps the most illuminating result with narrow-band stimuli was GY’s inability, when the abrupt onset and offset were removed, to discriminate between red and blue or red and blank, which is consistent with the old notion that the superior colliculi are especially effective at detecting transients. There is an important rider to the arguments above. It is known that the sensitivity of the M and L-cones is sufficiently broad for them to respond weakly to wavelengths as short as 426 nm (e.g. Stockman & Sharpe, 1999), a fact not obvious from the customary normalised sensitivities shown in Fig. 7. However, although the L and M-cones do have a quantum catch right
Table 3 Cone excitation for the stimuli used in experiment 4. Luminance
Y
x
y
L-cone
S-cone
Performance
Red 125 150 175 75 200 255 100 Background
1.81 8.15 13.9 18.3 2.51 26.8 46.2 5.47 4.6
0.583 0.285 0.289 0.284 0.29 0.286 0.284 0.303 0.29
0.357 0.6 0.595 0.597 0.592 0.595 0.594 0.571 0.325
0.813419 0.649824 0.651305 0.649689 0.651781 0.650401 0.649853 0.657124 0.679429
0.155507 0.165661 0.169145 0.173496 0.173649 0.174235 0.179707 0.19584 1.154527
ns ns s s s s s
532
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
across the visible spectrum their relative sensitivity is so low in the short wavelength range that it can hardly account for why in blindsight the patient is actually better at detecting narrow-band blue than green stimuli. Similarly it is difficult to see why, for GY, narrow-band red was so good. Furthermore, even if the L and M-cones could detect the narrow-band blue stimuli in the normal hemiretina any effectiveness in a degenerated and blindsighted hemiretina is unknown. Unfortunately there are no single cell recordings in the mid-brain of monkeys in which V1 has been removed. The general message from the set of experiments on GY and MS is that although there are circumstances where they can localise, and discriminate between, a variety of chromatic and achromatic stimuli in their blind hemifields, their highly successful performance is based on relatively simple aspects of the stimuli, notably sharp contrast borders, relative intensity, and steep temporal onsets and offsets. All these features allow the subjects to detect ‘events’ but the events appear to vary only in subjective salience (here meaning roughly strength of feeling, or valency) and when salience is equated discrimination becomes impossible. With the stimuli used here categorisation of colours was absent when GY was asked to identify them (experiment 4) but, importantly, some chromatic processes are intact in blindsight as revealed by the detection task in experiment 2. This notion of the simple nature and importance of salience was first clearly expressed by Humphrey (1974) but has been curiously neglected in subsequent work on blindsight. Of great interest in this respect is the study by Morland et al. (1999), which demonstrated that GY could successfully discriminate and match colours, but not brightness, between hemifields. Taken together our experiments provide evidence that it is certain simple stimulus attributes that remain intact in blindsight but the role of the two channels with distinct spatio-temporal properties mediating blindsight, as first revealed by Barbur et al. (1994), remains to be established. Finally the experiments, especially when taken in conjunction with the finding that uncertainty about the precise timing of an expected stimulus drastically reduces its detection (Cowey & Stoerig, 2003) even though the stimulus has sharp edges and steep temporal gradient at onset and offset, indicate that even in the most intensively studied subject GY, blindsight might be of little practical use in everyday life and that attempts at rehabilitation might best be directed at training hemianopes to use their surviving visual field more effectively. Acknowledgment This research was supported by a UK Medical Research Council Grant to AC and an EPA Cephalosporin Trust Award to IA. References Azzopardi, P., & Cowey, A. (2001). Motion discrimination in cortically blind patients. Brain, 124, 30–46. Barbur, J. L., Harlow, A. J., & Weiskrantz, L. (1994). Spatial and temporal response properties of residual vision in a case of hemianopia. Philosophical Transactions of the Royal Society, London, B343, 157–166. Barbur, J. L., Ruddock, K. H., & Waterfield, V. A. (1980). Human visual responses in the absence of the geniculo-calcarine projection. Brain, 103, 905–928. Barbur, J. L., Watson, J. D. G., Frackowiak, R. S. J., & Zeki, S. (1993). Conscious visual perception without V1. Brain, 116, 1293–1302. Brent, P. J., Kennard, C., & Ruddock, K. H. (1994). Residual colour vision in a human hemianope: Spectral responses and colour discrimination. Proceedings of the Royal Society London, B, 25, 219–225. Bridge, H., Thomas, O., Jbabdi, S., & Cowey, A. (2008). Changes in connectivity after visual cortical brain damage underlie altered visual function. Brain, 131, 1433–1444. Cowey, A. (2004). Fact, artefact and myth about blindsight. Quarterly Journal of Experimental Psychology, 57A, 577–609. Cowey, A. (2010). The blindsight saga. Experimental Brain Research, 200, 3–24. Cowey, A., Alexander, I., Heywood, C., & Kentridge, R. (2008). Pupillary responses to coloured and contourless displays in total cerebral chromatopsia. Brain, 131, 2153–2160. Cowey, A., & Stoerig, P. (1991). The neurobiology of blindsight. Trends in Neuroscience, 14, 140–145. Cowey, A., & Stoerig, P. (1999). Spectral sensitivity in hemianopic macaque monkeys. European Journal of Neuroscience, 11, 2114–2120. Cowey, A., & Stoerig, P. (2003). Stimulus cueing in blindsight. In C. A. Heywood, A. D. Milner, & C. Blakemore (Eds.). The roots of visual awareness. Progress in brain research (Vol. 144, pp. 261–277). Amsterdam: Elsevier. Cowey, A., Stoerig, P., & Perry, V. H. (1989). Transneuronal retrograde degeneration of retinal ganglion cells after damage to striate cortex in macaque monkeys: Selective loss of P beta cells. Neuroscience, 29, 65–80. Cowey, A., & Walsh, V. (2000). Magnetically induced phosphenes in sighted, blind and blindsighted observers. NeuroReport, 11, 3269–3273. Dartnall, H. J. A., Bowmaker, J. K., & Mollon, J. D. (1983). Microspectrophotometry of human photoreceptors. In J. D. Mollon & L. T. Sharpe (Eds.), Colour vision physiology and psychophysics (pp. 69–80). New York: Academic Press. De Monasterio, F. M. (1978). Properties of ganglion cells with atypical receptive-field organisation in the retina of macaques. Journal of Neurophysiology, 41, 1435–1449. Hall, N. J., & Colby, C. L. (2009). Response to blue visual stimuli in the macaque superior colliculus. Society for Neuroscience [Abstracts, 756]. Heywood, C. A., & Cowey, A. (2003). Colour vision and its disturbances after cortical lesions. In M. Fahle & M. Greenlee (Eds.), The neuropsychology of vision (pp. 259–281). Oxford: Oxford University Press. Humphrey, N. K. (1974). Vision in a monkey without striate cortex: A case study. Perception, 3, 241–255. Keating, E. G. (1979). Rudimentary color vision in the monkey after removal of striate and preoccipital cortex. Brain Research, 179, 379–384. Klüver, H. (1942). Functional significance of the geniculo-striate system. Biological Symposia, 3, 253–299. Klüver, H. (1949). Visual functions after removal of the occipital lobes. Journal of Psychology, 11, 23–45. Leh, S. E., Mullen, K. T., & Ptito, P. (2006). Absence of S-cone input in human blindsight following hemispherectomy. European Journal of Neuroscience, 24, 2954–2960. Leporé, F., Cardu, B., Rasmussen, T., & Malmo, R. B. (1975). Rod and cone sensitivity in destriate monkeys. Brain Research, 93, 203–221. Malmo, R. B. (1966). Effects of striate cortex ablation on intensity discrimination and spectral intensity distribution in the rhesus monkey. Neuropsychologia, 4, 9–16. Marrocco, R. T., & Li, R. H. (1977). Monkey superior colliculus: Properties of single cells and their afferent inputs. Journal of Neurophysiology, 40, 844–860. Morland, A. B., Jones, S. R., Finlay, A. L., Deyzac, E., Le, S., & Kemp, S. (1999). Visual perception of motion, luminance and colour in a human hemianope. Brain, 122, 1183–1198. Ptito, A. (2007). Neural substrates of blindsight after hemispherectomy. Neuroscientist, 13, 506–518.
I. Alexander, A. Cowey / Consciousness and Cognition 19 (2010) 520–533
533
Schilder, P., Pasik, P., & Pasik, T. (1972). Extrageniculostriate vision in the monkey. III. Circle vs. triangle and ‘red vs. green’ discrimination. Experimental Brain Research, 14, 436–448. Schiller, P. H., & Malpeli, J. G. (1977). Properties and tectal projections of monkey retinal ganglion cells. Journal of Neurophysiology, 40, 428–445. Smithson, H. E., Sumner, P., & Mollon, J. D. (2003). How to find a tritan line? In J. D. Mollon, J. Pokorny, & K. Knoblauch (Eds.), Normal and defective colour vision (pp. 279–287). Oxford: Oxford University Press. Stockman, A., & Sharpe, L. T. (1999). Cone spectral sensitivities and color matching. In K. R. Gegenfurtner & L. T. Sharpe (Eds.), Color vision (pp. 53–87). Cambridge: Cambridge University Press. Stoerig, P. (1985). Chromaticity and achromaticity: Evidence for a functional differentiation in visual field defects. Brain, 110, 869–886. Stoerig, P. (2006). Blindsight, conscious vision, and the role of primary visual cortex. Progress in Brain Research, 155, 217–234. Stoerig, P., & Cowey, A. (1989). Wavelength sensitivity in blindsight. Nature, 342, 916–918. Stoerig, P., & Cowey, A. (1992). Wavelength discrimination in blindsight. Brain, 115, 425–444. Stoerig, P., & Cowey, A. (1997). Blindsight in man and monkey. Brain, 120, 535–559. Sumner, P., Adamjee, T., & Mollon, J. D. (2002). Signals invisible to the collicular and magnocellular pathways can capture visual attention. Current Biology, 12, 1312–1316. Weiskrantz, L. (1998). Consciousness and commentaries. Towards a science of consciousness II – The second Tucson discussions and debates. Cambridge: MIT Press.
Consciousness and Cognition 19 (2010) 534–546
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Feeling of doing in obsessive–compulsive checking S. Belayachi a,*, M. Van der Linden a,b a b
Cognitive Psychopathology Unit, University of Liège, Belgium Cognitive Psychopathology and Neuropsychology Unit, University of Geneva, Switzerland
a r t i c l e
i n f o
Article history: Received 22 June 2009 Available online 23 February 2010 Keywords: Obsessive–Compulsive Disorder Checking Sense of agency Feeling of doing Action-monitoring
a b s t r a c t Research on self-agency emphasizes the importance of a comparing mechanism, which scans for a match between anticipated and actual outcomes, in the subjective experience of doing. This study explored the ‘‘feeling of doing” in individuals with checking symptoms by examining the mechanism involved in the experienced agency for outcomes that matched expectations. This mechanism was explored using a task in which the subliminal priming of potential action-effects (emulating outcome anticipation) generally enhances people’s feeling of causing these effects when they occur, due to the unconscious perception of a match between primed and observed outcomes. The main result revealed a negative relationship between checking and self-agency for observed outcomes that were primed prior to actions. This suggests that checking individuals fail to grasp the correspondence between actual outcomes of their actions and expected ones. We discuss the possible role of undermined self-agency in checking phenomena and its relationship with cognitive dysfunction. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Obsessive–Compulsive Disorder (OCD) is characterized by both obsessions (i.e., recurrent thoughts or images, particularly ones that cause distress) and compulsions (i.e., urges to perform mental or physical acts repeatedly). A fundamental characteristic of the OCD phenomenology is the peculiar way in which individuals with those symptoms construe what they are doing in certain behavioral situations. Current cognitive models suggest that compulsive behaviors are likely to be accompanied either by the aim of preventing threatening events (i.e., harm avoidance; Rachman, 1997; Salkovskis, 1985) or by a particular sensation ‘‘that something’s wrong” with an action or the environment (incompleteness; Coles, Frost, Heimberg, & Rhéaume, 2003). Interestingly, the prevalence of these two core dimensions seems to vary across the different subtypes of OCD: harm avoidance may particularly characterize washing and obsessing symptoms (Tolin, Brady, & Hannan, 2008), while incompleteness may be especially associated with checking (Coles et al., 2003; Tolin et al., 2008) and ordering (Ecker & Gönner, 2008). Incompleteness and ‘‘not quite right” feelings are characterized by impressions of failure or imperfection, which can lead to an inability to achieve a sense of ‘‘task completion” or ‘‘closure” regarding actions (e.g., locking the door) or perceptions (e.g., objects on a table). In the particular case of action incompleteness (i.e., checking behaviors), people may experience inconsistent feelings that ‘‘actions or intentions have been incompletely achieved” (Summerfeldt, Huta, & Swinson, 1998, p. 80), or feel only a weak sense of goal satisfaction. Such conflicting appraisals of one’s behaviors have been related to impairments of the ability to monitor actions (Fitzgerald et al., 2005; Gehring, Himle, & Nisenson, 2000; Maltby, Tolin,
* Corresponding author. Address: Cognitive Psychopathology Unit, Department of Cognitive Sciences, University of Liège, Belgium. Fax: +32 4 366 28 08. E-mail address:
[email protected] (S. Belayachi). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.02.001
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
535
Worhunsky, O’Keefe, & Kiehl, 2005; Pitman, 1987; Ursu, Stenger, Shear, Jones, & Carter, 2003). For example, Pitman (1987) proposed that compulsive behaviors may stem from a recurrent perception of a ‘‘mismatch signal” informing one of a discrepancy between actual outcomes and intended effects; checking symptoms have been consistently connected with abnormal action-monitoring (Hajcak & Simons, 2002). The role of action-monitoring is to assess the extent to which ongoing actions and subsequent outcomes match up to what was initially intended (i.e., internal goals and plans), and to trigger either a matching signal (i.e., goal satisfaction) or a mismatch signal, when goals may not be attained or expected effects may not be obtained (Aarts, Custers, & Wegner, 2005; Pitman, 1987; Ridderinkhof, van den Wildenberg, Segalowitz, & Carter, 2004). In addition, by triggering those informing signals, action-monitoring plays an important role in the subjective understanding of ‘‘what one is doing” or ‘‘what one has just done”; a dysfunction at this level may thus lead to an inconsistent appraisal of one’s action. More concretely, those matching signals may generate an internal feeling of doing; this ‘‘impression” would in turn constitute the phenomenal cue of goal attainment, on the basis of which people are able to naturally end their action (Aarts et al., 2005; Woody et al., 2005). Hence, an inability to generate a matching signal when observed effects correspond to expected ones may undermine the experience of having actually performed those intended effects. From this perspective, lacking such a basic ‘‘feeling of doing” could make people less sure of whether they have actually satisfied a goal and cause them to feel incompleteness, which may in turn trigger adjustment behaviors to ensure that the goal has been completed (i.e., checking behaviors). The present study therefore aimed to explore checking individuals’ ability to generate a ‘‘feeling of doing” for outcomes that obviously match expected effects. Theories of the sense of agency (i.e., the subjective experience of doing) offer a reliable theoretical context for understanding how such a basic feeling of doing can be modulated by the unconscious perception of a correspondence between observed action-effects and expected ones (Aarts, Wegner, & Dijksterhuis, 2006; Aarts et al., 2005; Wegner & Wheatley, 1999). Experience of agency may arise when one feels that a perceived outcome (e.g., a locked door) is caused by one’s action (e.g., putting a key in the lock), and that both actions and their related effects correspond to mental representations (e.g., the desire to secure the house), which are generally intentions, goals and expected outcomes (Aarts et al., 2005, 2006; Wegner, 2002; Wegner & Sparrow, 2004; Wegner & Wheatley, 1999).The perception of a match between perceived outcomes and anticipated ones would trigger the basic feeling that we had caused some intended effects through our actions. Accordingly, people may feel that they have caused specific effects whenever these effects are mentally anticipated (Wegner & Sparrow, 2004); so long as an unconscious mechanism is capable to grasp a correspondence between anticipated and observed effects and subsequently to trigger a matching signal (Aarts et al., 2005, 2006). In fact, research on the sense of agency has demonstrated that the experience of agency can be artificially induced in tasks in which participants’ behavioral outcomes coincide with prior consistent thoughts about those outcomes. For example, Aarts et al. (2005) demonstrated that priming action-effects just prior to participants’ actions enhanced the feeling that they had caused those effects when they actually occurred. Their work was based on the assumption that subliminal priming mimics the automatic activation of the representations of action-effects before the action, while simultaneously preventing conscious awareness of these thoughts. In their study, the authors used a task in which the participant and the computer each moved a square in opposite directions; the participants’ task was to press a key (i.e., move) to stop the motion of the squares (i.e., ‘‘actual outcome”), and subsequently to determine whether they or the computer could have caused the square to stop in the observed position (i.e., agency judgment). In reality, the participants did not have any control over causing the square to stop in that particular position, which was arranged so that it never represented either the participants’ or the computer’s real stop position. In half of the trials, the square position to be presented was primed just before the participants stopped the motion of the square (i.e., prior thoughts on expected effect), and the results showed that effect-priming significantly enhanced the ‘‘feeling of personally causing” the square to stop in the presented position. In this priming task, people’s illusion of agency is thought to reflect the automatic comparison between observed and anticipated effects, leading to a matching signal that may be responsible for the increased impression of having caused the observed outcomes. Thus, this task allows the assessment of the ability to grasp a correspondence between observed and expected effects, on the basis of which one may subsequently experience a feeling of agency for those effects. To sum up, the perception of a match between expected effects and observed effects of actions tends to make people more prone to experience a feeling of doing for those effects (Aarts et al., 2006; Wegner & Sparrow, 2004). Furthermore, this mechanism is claimed to be responsible for the illusion of agency observed in experimental settings whenever perceived outcomes are primed just before the participants act (Aarts et al., 2005, 2006; Jones, de-Wit, Fernyhough, & Meins, 2008). Consequently, an altered comparing process, leading to an inconsistent matching signal, should lead people to experience less of a ‘‘feeling of doing” in the agency judgment task developed by Aarts et al. (2005). Consistently, checking symptoms seem to be specifically related to abnormal action-monitoring, which may lead to inconsistent mismatch signals, in addition to feelings of incompleteness (e.g., Hajcak & Simons, 2002). Furthermore, it has been proposed that individuals’ feelings of having an unfulfilled goal, which characterize checking, may be related to the inability to generate an internal signal that normally triggers a basic sense of task completion (Szechtman & Woody, 2004). From this perspective, individuals with checking symptoms may have a comparing mechanism that is less efficient at grasping an accurate match between real outcomes and expected ones, which in turn leads checkers to lack any feeling of doing for those outcomes. The methodology of Aarts et al. (2005) allows one to examine this issue by means of the level of experienced agency for outcomes that have been primed prior to participants’ action (i.e., in the priming of the outcome condition of the agency judgment task). Indeed, the illusion of agency in this condition has been interpreted as resulting from the perception of a match between primed and observed outcomes, and the generation of a matching signal by the
536
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
comparing mechanism. Considering that a defective comparing mechanism would undermine experienced agency in the priming condition, and if checking symptoms are indeed connected with such an impairment, we could expect that those symptoms would be related to lower experienced agency in the priming condition of the agency task. Since ‘‘not just right” experiences in ordering symptoms may concern perception rather than action, and the other OCD dimensions seem not to be phenomenologically related to the action component explored in this study, we expected that checking symptoms appear specifically associated with lower experience of agency for observed outcomes that match ‘‘anticipated” ones (because of subliminal priming), as compared to the other OCD dimensions. Those issues will be examined in individuals with subclinical levels of OCD symptoms, as non-clinical participants with checking proneness have been found to have clinical features and cognitive impairments similar to those identified in checking patients (Muris, Merckelbach, & Clavan, 1997; Zermatten, Van der Linden, Larøi, et al., 2006).
2. Method 2.1. Participants One hundred and twenty-eight undergraduates from the University of Liège, aged between 18 and 36, participated in the study. Participants were randomly recruited from various faculties and schools of the university (i.e., Law and Political Science, Psychology and Education, Science, Applied Science, Management, Social Sciences); they were not compensated for their participation. Data from three participants were omitted because they did not respect the instructions. In addition, the debriefing revealed that 16 participants may have realized the true nature of the study during the task (see the debriefing description in Section 2.4.); those participants’ data were excluded from the analysis. Data from two participants also had to be discarded as they appeared to be critical outliers (cf. Section 3). The reported results are from the remaining 107 participants (39 males and 68 females). Their mean age was 22.5 years (SD = 3.14 years). 2.2. Experimental task and procedure Informed consent was obtained from all participants following a full explanation of the experimental procedure. Detailed written and oral instructions explained that we were interested in people’s feelings of control and that participants would be asked to direct the movement of a black square on a screen. In an individual testing session, participants performed the illusion of agency task (i.e., the Wheel of Fortune Task) and completed two questionnaires assessing OCD and depression symptoms, as well as questionnaires unrelated to the present study. Half of the participants first undertook the Wheel of Fortune Task, and then completed all the questionnaires; the other half completed the questionnaires before the task. The order of the questionnaires was counterbalanced across participants. Participants were debriefed at the end of the testing session. The illusion of agency task was programmed in the Matlab environment; we used the same task administration and scoring described in Aarts et al. (2005). The task consisted of 16 trials in which participants first had to press down and hold the S key of the keyboard to cause a gray square to move rapidly in a rectangular path in a counter-clockwise direction. At the same time, another gray square (‘‘under the computer’s control”) moved along the same path but in a clockwise direction. At a random point in time, the gray squares could no longer be seen on the screen and participants had to press the Enter key to stop the motion of the squares they could no longer see but which were said to still be rotating. Immediately after the Enter key press, a black square appeared in a specific location on the path (i.e., representing the potential outcome of the participants’ action of pressing the Enter key), and participants had to indicate whether they had caused the square to stop in that position. In eight trials (priming of outcomes condition), the black square’s position was primed just before the participants pressed the Enter key, while there was no priming of effects in the other eight trials (no priming of outcomes condition). The order of trials with and without priming of potential outcomes was randomly specified (for each participant, the computer randomly determined the order of the 16 trials at the beginning of the task). The different events in each trial were as follows: 2.2.1. Setting the square in motion Each trial began with a 3-s warning signal (the word ‘‘Warning” in French appeared on the screen); then, the message ‘‘Start” appeared in the center of a rectangular path, made out of eight white squares, until the participant pressed the S key. The participant’s and the computer’s squares then began to move along the path in alternate steps (the squares were presented one after the other). In fact, the white squares representing the eight different positions on the path turn gray whenever one of the squares passes through them; each gray square was displayed for 60 ms in each position. Thus, it took 960 ms for one complete lap to occur (60 ms 8 positions 2 [participant’s and computer’s] squares; Aarts et al., 2005). The number of laps that occurred in a trial was programmed to vary randomly between 8 and 10. In each condition, the starting location of the squares varied among all possible locations throughout the eight trials, and each possible starting position was randomly selected. 2.2.2. Priming event In the priming of outcomes condition, the black square that was to be displayed (i.e., potentially representing the participants’ action-effect) was flashed on the screen before the appearance of the message ‘‘Stop.” The subliminal priming of
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
537
the black square position occurred 40 ms after the last presentation of the participant’s square, for 34 ms; 46 ms later, it was followed by the message ‘‘stop.” The primed location was always the same as the subsequently presented position of the black square. In the no priming of outcomes condition, the position of the black square was not flashed (the position was presented in white the whole time). The priming event was employed for every possible location, resulting in eight replications of each condition. 2.2.3. Stopping the square At a random point in time (between 8 and 10 laps of the path), the word ‘‘Stop” appeared in the center of the grid until the participant pressed the Enter key to stop the motion of the square. When the message ‘‘Stop” appeared on the screen, only the eight empty white tiles were shown: the gray squares could no longer be seen but participants were told that the squares were still continuing to move around the path and that they had to stop the progress of their square. The participants’ action (pressing the Stop button) resulted in one of the white squares of the grid turning black, which was said to represent the position of either their own square, or the computer’s, at the time they pressed Stop. This black square appeared 100 ms after the Enter key press, and was presented for 1 s. The location of this square was always four positions farther than the last position of the participant’s square before the Stop message had appeared, so that they had no actual control over where the black square landed. In each condition (the priming of outcomes and no priming of outcomes conditions), this stop location occurred in each of the eight tiles of the path. 2.2.4. Agency ratings Immediately after the presentation of the black square, participants were asked to indicate whether they felt they had caused the square to stop on that position or whether the computer had. Agency ratings were made on a 10-point Likert scale that appeared on the screen, running from ‘‘not me at all” (0) to ‘‘absolutely me” (9). Participants were told that a ‘‘not me” response was represented by values from 0 (if they were certain) to 4 (if they were not sure), while points from 5 to 9 represented a ‘‘me” response, with 5 if they were not sure, and 9 if they were certain. They were also asked to use intermediate values to moderate their judgment and to rely on their feelings if they had difficulties generating a response. 2.3. Measures 2.3.1. Experienced agency For reasons of ease, participants were asked to rate their feelings of causing the square to stop in that position by using the 0–9 keys of the number pad; these agency ratings were subsequently rescaled on a 1–10 scale, for each of the 16 trials. Mean ratings of experienced agency were calculated for both the priming of outcomes condition and for the no priming of outcomes condition. 2.3.2. Priming sensitivity index As agency ratings in the priming and no-priming conditions provide information only about the extent to which participants self-attributed the origin of outcomes; we also used the difference in agency ratings between the two conditions to quantify the amount of the increase in experienced agency due to the prior priming of potential outcomes (i.e., the effect of priming of outcome). Thus, for each participant, we computed a priming sensitivity index, defined as the mean agency ratings in primed trials less mean agency ratings in unprimed trials. Mean scores on the priming sensitivity index may range from 9 to 9. 2.3.3. Potential control We used the participants’ reaction times (in ms) between the Stop message and the Enter key press (i.e., stopping the square) as a measure of the probability that they had actually stopped their ‘‘invisible but still rotating square” in the same position as that indicated by the black square (i.e., potential control ‘‘over producing the outcomes”). As Aarts et al. (2005) mentioned, it would take 330 ms after the presentation of the Stop message to land on the position of the black square; thus, if the participants pressed the Enter key 330 ms after the presentation of the Stop message, they would have landed on the position of the black square (see Aarts et al., 2005, for further explanation of this measure). For each participant, we calculated a measure of potential control, defined as the absolute difference between the response time after the Stop message and the time required to land exactly on the position of the black square (330 ms). The smaller the difference, the more likely that they actually could have caused the square to land on that position (see Aarts et al., 2005). Trials in which reaction times were greater than 960 ms (i.e., if the participants pressed the Enter key after the invisible rotating square had started to make a new lap since the last presentation of their gray square) were first corrected, by subtracting 960 ms (i.e., the time for a complete lap to occur) from those slower responses, then the difference between this new response time and 330 ms was calculated (Aarts et al., 2005). Trials in which participants’ reaction times were greater than 1920 ms were removed from the data (Jones et al., 2008); they represented 3.9% of the total trials in the whole sample. 2.3.4. Obsessive–Compulsive Inventory – Revised (OCI-R, Foa et al., 2002) The validated French version of the OCI-R (Zermatten, Van der Linden, Jermann, et al., 2006) is a self-report questionnaire that consists of 18 items evaluating OCD symptoms and is composed of six subscales: ‘‘Washing,” ‘‘Obsessing,” ‘‘Ordering,”
538
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
‘‘Checking,” ‘‘Hoarding,” and ‘‘Neutralizing.” Participants were asked to determine to what extent the situation described in each particular statement had distressed them during the past month, using a 5-point scale (0 = ‘‘not at all”; 4 = ‘‘extremely”). Total scores range from 0 to 72; the OCI-R subscales scores both range from 0 to 12. Cronbach’s alphas indicated good to acceptable internal consistency for all the measures (OCI-R total score: .81; Checking: .77; Washing: .72; Obsessing: .71; Ordering: .84; Hoarding: .66; Neutralizing: .52). The range of scores obtained on each OCD measure included scores comparable to those observed in clinical samples (OCI-R total score: 1–46; Checking: 0–10; Washing: 0–12; Obsessing: 0–10; Ordering: 0–11; Hoarding: 0–11), except for the Neutralizing OCI-R subscale (0–7). Mean scores for OCD dimensions are shown in Table 1 and revealed that Ordering symptoms were the most frequent, while Washing dimensions were less frequently observed in a non-clinical sample, replicating previous studies of OCD symptoms in the general population (Zermatten & Van der Linden, 2008a). However, statistical tests of homogeneity revealed that the mean Washing, Obsessing, Hoarding and OCI-R total scores are higher in our study than in a French-speaking sample of 220 undergraduates in Zermatten and Van der Linden’s (2008a) (ps < .01); whereas mean Neutralizing, Ordering and Checking scores were comparable across the two studies (ps > .25). There was no gender difference for all those measures (ps > .08). 2.3.5. Center for Epidemiological Studies Depression Scale (CES-D, Radloff, 1977) The French adaptation of the CES-D (Fuhrer & Rouillon, 1989) was used to assess participants’ proneness to depression. This self-report measure consists of 20 items describing states of happiness or depressed mood. For each item, participants had to rate the extent to which this situation applied to them during the last week, by using a 4-point Likert scale, ranging from 0 = ‘‘Never, rarely (less than 1 day)” to 3 = ‘‘Frequently, all the time (from 5 to 7 days).” Responses to items that described states of well-being were reversed; total scores range from 0 to 60. The French version of the CES-D has good overall psychometric properties and a factorial structure similar to that observed in the original English version (Fuhrer & Rouillon, 1989). As indicated by Cronbach’s alpha (.89), this scale has a good internal consistency. A statistical test of homogeneity confirmed that the mean score on the CES-D obtained in this study (see Table 1) did not differ from that observed in a French-speaking sample of 291 undergraduates in Riddle, Blais, and Hess’s (2002) study (p > .52). There was no gender difference (p > .13). 2.4. Debriefing Following the procedure in the Aarts et al. (2005) study, participants were debriefed after the completion of the entire protocol. The debriefing began by asking the participants whether they had ‘‘seen something special during the task,” in order to ensure that primed locations were not perceived. We noted any response that referred to the perception of a flash or another square before the stop message; those participants were treated as having perceived the prime. Participants’ answers were followed by the information that the presented black square sometimes appeared very quickly on the screen just before the stop message. This information was followed by a question about whether they had seen it or not, depending on the participants’ reaction (e.g., an assenting reaction was followed by the question: ‘‘you did notice it?”, while the question ‘‘you did not notice it?” was asked if participants had no reaction or were surprised). Participants who did not explicitly report having seen the black square flash in response to the first question but admitted in response to the second that they had seen it at least once were treated as having perceived the prime. The debriefing indicated that six participants had realized the true nature of the study because the paradigm was presented in social psychology courses they had taken. In addition, 10 participants reported having seen a black square flash (subliminal priming) in at least one trial and a few of them noted that the position of this square was identical to the location of the final square. Those participants’ data were excluded from the analysis (see above), as their experience of the task was clearly not comparable to that of the other participants.
Table 1 Mean scores and SDs for the experienced agency (priming and no priming of outcome conditions), priming sensitivity, OCD, and depression measures for the whole sample and for each checking group. Dependent variables Agency task Priming of outcome condition No priming of outcome condition Priming sensitivity index OCD dimensions OCI-R checking OCI-R washing OCI-R obsessing OCI-R hoarding OCI-R ordering OCI-R neutralizing OCI-R total score Depression Note: SD: standard deviation.
Whole sample score (SD) 5.1 (1.1) 4.6 (1.2) 1.1 (0.9) 2.2 1.7 3.0 4.0 3.8 1.0 15.7 15.8
(2.1) (2.2) (2.5) (2.6) (2.9) (1.5) (8.5) (9.3)
Non-checking group 5.6 (1.12) 4.8 (1.42) 0.88 (1.32) 0 1.1 1.9 2.4 2.2 0.4 8.3 12.2
(2.26) (2.25) (1.65) (2.12) (1.34) (5.66) (8.67)
Checking group 4.6 (0.83) 4.4 (0.96) 0.27 (1.13) 5.7 2.8 3.4 5.2 5.8 1.7 24.5 18.4
(1.70) (2.71) (2.36) (1.92) (2.71) (1.98) (7.31) (9.68)
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
539
3. Results The six OCI-R subscales were not normally distributed; those data were transformed using the Box–Cox transformation to comply with normality and homogeneity of variance. Multivariate outliers were checked using the Cook’s distance index. Fox (1991) proposed to define multivariate outliers as cases that have a Cook’s distance greater than the cutoff 4/(n k 1) (where n is the number of cases and k is the number of predictors). Preliminary analysis identified two participants as multivariate outliers, with a Cook’s distance >0.04 (i.e., 4/[109 8 1]; we decided to exclude data from these two participants. 3.1. Preliminary analysis 3.1.1. Effect of outcome priming on experienced agency in the whole sample Mean ratings of experienced agency were calculated for both the priming of outcomes condition and the no priming of outcomes condition. The sample’s mean scores for those measures are shown in Table 1. Statistical homogeneity tests confirmed that the mean agency judgments on primed trials and on non-primed trials obtained in this study did not differ from those observed in a sample of 143 undergraduates in the Jones et al. (2008) study (ps > .20); they were also similar to that observed in the Aarts et al. (2005) sample of 50 participants (ps > .06). Because Jones et al. (2008) had found a gender effect on the illusion of agency – priming of outcomes enhanced feelings of agency in women but not in men – we performed a 2 (Outcome: Primed vs. Unprimed) within-participants 2 (Gender: male vs. female) between-participants ANOVA. Effect size was tested through the f estimate, as suggested by Cohen (1988); according to Cohen’s rule of thumb, an effect size of .10 is a small effect, .25 a medium effect and .40 a large effect. For each effect, the proportion of variance explained was left in parentheses (g2). This analysis revealed a significant main effect of priming, suggesting that participants’ agency feelings for observed outcomes were higher when they were primed just before their actions than when they were not primed at all, F(1, 105) = 9.79, p = .002, f = .31 (g2 = .09). This effect size is medium to large by Cohen’s (1988) criteria and is comparable to the effect sizes for effect-priming of g2 = .09 found by Aarts et al. (2006) and g2 = .05 reported by Jones et al. (2008). As in the Jones et al. (2008) study, there was no main effect of gender, F(1, 105) = 2.03, p = .157, while the interaction between priming and gender was found to be significant, F(1, 105) = 6.45, p = .013, f = .25 (g2 = .06). By Cohen’s (1988) criteria, this is a small to medium effect, and is less comparable to the effect size for effect-priming of g2 = .11 found by Jones et al. (2008). Paired t-tests performed separately for each gender on agency ratings across the two conditions revealed that there was no effect of priming for men, t(38) = 0.33, p = .742); for women, agency ratings were significantly higher for the priming of outcome than the no priming of outcome trials, t(67) = 5.05, p < .001, Cohen’s d = .70. This effect size is comparable, even higher, to that observed in the Jones et al. study (Cohen’s d = .59). As well, t-tests performed for agency ratings on the primed and non-primed trials and the priming sensitivity index across genders showed that there were no gender differences on primed trials, t(105) = 0.34, p = .732, whereas men scored significantly higher than women on non-primed trials, t(105) = 2.58, p = .011, Cohen’s d = .50; this result is comparable to that observed in the Jones et al. study (Cohen’s d = .52). In addition, women scored significantly higher than men on the priming sensitivity index t(105) = 2.54, p = .013, Cohen’s d = .49. 3.1.2. Potential control over producing the effect For each trial, we calculated the absolute difference between the response time after the message to stop and the initial time required to land exactly on the position of the black square. The mean absolute difference scores for the outcomeprimed condition and the unprimed condition were then calculated and entered in a 2 (Outcome: primed, unprimed) 2 (Gender: male, female) mixed-design ANOVA. This analysis was conducted on Log-transformed means, because equal variances were not assumed for both the outcome-primed condition and the unprimed condition (based on Levene’s tests of homogeneity of variance). This analysis revealed no effect of priming, F(1, 105) = 2.34, p = .129, indicating that priming of outcome did not affect participants’ potential control over producing the effect; the mean absolute difference score was 70 ms (SD = 81). There was a main effect of gender, F(1, 105) = 4.55, p = .035, f = .20 (g2 = .04), suggesting that men were more likely to stop their square closer to the final position of the black square on both kinds of trials. Finally, there was no interaction between gender and priming, F(1, 105) = 1.53, p = .218.
3.2. OCD symptoms and feeling of doing 3.2.1. Correlation analyses We first calculated Pearson correlations for the priming sensitivity index and agency ratings in the priming and in the non-priming conditions with the measures of OCD (i.e., the six transformed OCI-R subscales and OCI-R total scores) and depression (i.e., CESD). To correct for multiple comparisons, a Bonferroni correction of (0.05/24) = 0.002 was used. Those analyses are summarized in Table 2. Consistent with our hypothesis, checking symptoms were found to be negatively associated with agency ratings in the outcome-primed condition, but only at an uncorrected significance (p = .011). However, contrary to our hypothesis, there was no significant relationship between the overall benefit of priming and checking
540
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
Table 2 Pearson correlations between OCD dimensions and experienced agency (priming of outcome condition), priming sensitivity index and depression. Experienced agency (priming of outcome condition)
Experienced agency (no priming of outcome condition)
Priming sensitivity index
OCD dimensions OCI-R checking OCI-R washing OCI-R obsessing OCI-R hoarding OCI-R ordering OCI-R neutralizing OCI-R total score
0.24* 0.01 0.02 0.02 0.10 0.06 0.12
0.27** 0.14 0.11 0.15 0.09 0.23* 0.01
0.04 0.13 0.11 0.11 0.15 0.15 0.09
Depression
0.05
0.05
0.08
Note: There was no correlation significant at p < .002 (Bonferroni correction). Indicates uncorrected significance at p < .05. ** Indicates uncorrected significance at p < .01. *
symptoms. In addition, checking and neutralizing symptoms negatively correlated with agency ratings in the non-priming condition at an uncorrected significance (ps > .005). 3.2.2. Regression analyses Considering the potentially confounding influences of correlations between symptoms (OCD subtypes and depression), zero-order correlations alone cannot determine the extent to which checking is specifically related to lower experienced agency, as compared to the other OCD dimensions. Linear regression analyses were therefore conducted to examine whether checking symptoms predicted lower experienced agency, when the other OCD dimensions and depression are statistically controlled. Because the preliminary analysis showed an effect of gender on some agency measures, gender was also controlled in each regression analysis; we set women at 1 and men at +1, so that a positive association corresponds to men. Thus, three multiple regression analyses were conducted to evaluate the specific associations between OCD dimensions and agency measures, when other OCD dimensions, depression and gender were partialled out. The dependent variables were the agency ratings in the priming condition, agency ratings in the non-priming condition and the overall benefit of priming; the six OCD dimensions, depression and gender were simultaneously entered as independent variables. Visual examination of the normal probability plot of the residuals and a Kolmogorov–Smirnov test for normal distribution of the standardized residuals suggested that there was no problem of heteroscedasticity for any of the regression analyses. Multicollinearity problems were checked by means of tolerance values (ranging from .66 to .88) and VIF values (ranging from 1.13 to 1.50). Those values suggest that there was no sign of multicollinearity, since VIF values over 2.5 and tolerance below .40 are considered as problematic (e.g., Allison, 1999). All the regression analyses are summarized in Table 3; a Bonferroni-corrected level of .017 (.05/3) was used. Effect size was tested through the f2 estimate, as suggested by Cohen (1988); according to Cohen’s rule of thumb, an effect size of .02 is a small effect, .15 a medium effect and .35 a large effect. The results indicate that lower experienced agency in the outcomeprimed condition was predicted only by checking symptoms (t = 2.49, p = .014, b = .29, f2 = 0.06), which according to Cohen’s criteria represents small to medium effect. The overall benefit of priming was marginally predicted by gender (t = 2.24, p = .027, b = .22, f2 = 0.05) and by ordering symptoms (t = 2.09, p = .039, b = .22, f2 = 0.04), both in the reverse direction. According to Cohen, these two effects are small to medium. Finally, agency ratings in the non-priming condition was predicted by gender (t = 2.53, p = .013, b = .23, f2 = 0.07), and by ordering (t = 2.33, p = .022, b = .23, f2 = 0.06) and hoarding symptoms (t = 2.18, p = .031, b = .20, f2 = 0.05), at an uncorrected significance. All these three observed effects are small to medium. In addition, checking symptoms also predicted agency ratings in the non-priming condition (t = 3.33, p = .001, b = .35, f2 = 0.12), but in the reverse direction and this effect size is small to medium. 3.2.3. Group comparison To address more directly the question of whether checking individuals, as compared to non-checking participants, are characterized by a lower illusion of agency (i.e., by a lesser priming effect or none at all), two groups were created based on participants’ checking subscores on the OCI-R. The checking group consisted of individuals whose scores fell within the top quartile of the distribution (score > 3; N = 22; six males and 16 females), and the non-checking group consisted of individuals with scores in the lowest quartile of the distribution (score = 0; N = 31; nine males and 22 females). Statistical homogeneity tests (ps > .30) suggest that the OCI-R checking scores of our checking group (M = 5.68; SD = 1.70) were comparable to the OCI-R checking scores of a non-clinical subgroup of checkers (5.60; SD = 1.96; Zermatten & Van der Linden, 2008a) and of a clinical population of OCD patients (M = 4.83; SD = 3.86; Foa et al., 2002). The mean scores of each group on the measures of agency, OCD and depression are shown in Table 1. In order to examine the effect of priming of outcome on experienced agency according to the level of checking symptoms, we performed a 2 (Outcome: Primed vs. Unprimed) within-participants 2 (Group: Checking vs. Non-checking) betweenparticipants ANCOVA. The gender was used as a covariate, with women set at 1 and men at +1. Effect size was tested
541
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
Table 3 Standardized, unstandardized regression coefficients, t, p values, partial correlation and effect size for gender, depression, and the six OCI-R subscales regressed on agency ratings in the priming condition, agency ratings in the no-priming condition and the priming sensitivity index. B
SE B
b
t
p
Partial r
f2
0.04 0.01 0.86 0.23 0.03 0.15 0.03 0.20
0.22 0.03 0.35 0.45 0.24 0.16 0.16 0.80
0.02 0.03 0.29 0.06 0.01 0.10 0.02 0.03
0.20 0.21 2.48 0.52 0.11 0.96 0.19 0.25
0.840 0.831 0.015 0.605 0.915 0.337 0.847 0.806
0.02 0.02 0.24 0.05 0.01 0.10 0.02 0.02
0.00 0.00 0.06 0.00 0.00 0.01 0.00 0.00
No priming of outcome condition Gender 0.36 depression 0.00 OCI-R checking 0.75 OCI-R washing 0.09 OCI-R obsessing 0.22 OCI-R hoarding 0.22 OCI-R ordering 0.24 OCI-R neutralizing 1.00
0.14 0.02 0.22 0.29 0.16 0.10 0.10 0.51
0.24 0.01 0.35 0.03 0.15 0.20 0.23 0.20
2.62 0.07 3.38 0.31 1.39 2.12 2.35 1.95
0.010 0.948 0.001 0.759 0.167 0.036 0.021 0.054
0.26 0.01 0.32 0.03 0.14 0.21 0.23 0.19
0.07 0.00 0.12 0.00 0.02 0.05 0.06 0.04
Priming sensitivity index Gender Depression OCI-R checking OCI-R washing OCI-R obsessing OCI-R hoarding OCI-R ordering OCI-R neutralizing
0.14 0.02 0.22 0.29 0.16 0.10 0.10 0.51
0.22 0.02 0.07 0.08 0.14 0.09 0.22 0.21
2.24 0.15 0.62 0.77 1.20 0.91 2.09 1.93
0.027 0.879 0.535 0.446 0.234 0.366 0.039 0.057
0.22 0.02 0.06 0.08 0.12 0.09 0.21 0.19
0.05 0.00 0.00 0.01 0.01 0.01 0.04 0.04
Priming of outcome condition Gender Depression OCI-R checking OCI-R washing OCI-R obsessing OCI-R hoarding OCI-R ordering OCI-R neutralizing
0.31 0.00 0.14 0.22 0.19 0.09 0.21 0.99
through the f estimate (adjusted for the addition of a covariate), as suggested by Cohen (1988); according to Cohen’s rule of thumb, an effect size of .10 is a small effect, .25 a medium effect and .40 a large effect. For each effect, the proportion of variance explained was left in parentheses g2p . This analysis revealed a significant main effect of priming, suggesting that participants’ agency feelings for observed outcomes were higher when they were primed just before their actions than when they were not primed at all, F(1, 50) = 4.38, p = .041, f = .29 g2p ¼ :08 . This effect size is medium to large by Cohen’s (1988) criteria. There was also a main effect of group, suggesting that checking participants’ reported lower agency ratings than participants with no checking symptoms F(1, 50) = 7.42, p = .009, f = .39 g2p ¼ :13 ; by Cohen’s (1988) criteria, this is a medium to large effect. There was an interac tion between priming and group, although it was not significant, F(1, 50) = 3.44, p = .070, f = .25 g2p ¼ :06 ; this is a small to medium effect (Cohen, 1988). One-way ANCOVAs conducted on agency ratings on the primed and non-primed trials across groups showed that there were no group differences on non-primed trials, F(1, 50) = 1.40, p = .241, whereas checking participants scored significantly lower than non-checking participants on primed trials, F(1, 50) = 12.67, p < .001, f = .50 g2p ¼ :20 ; this is a large effect by Cohen’s criteria. Finally, repeated-measures ANCOVAs were carried out separately for each group to examine agency ratings across the two conditions. These analyses revealed that Non-checking participants’ agency ratings were significantly higher for the priming of outcome than for the no priming of outcome trials, F(1, 29) = 5.83, p = .022, f = .42 g2p ¼ :17 , which is a large effect by Cohen’s criteria; the effect of priming failed to reach significance for the checking participants, F(1, 20) = 1.80, p = .19. 4. Discussion This study aimed to examine the ‘‘feeling of doing” in individuals with checking symptoms, by assessing their ability to experience agency for outcomes that matched expectations. Our results can be summarized as follows. First, and in line with previous findings (e.g., Aarts et al., 2005; Jones et al., 2008), priming the potential outcomes of participants’ actions generally enhanced experienced agency for those outcomes when they actually occurred (in the whole sample). Furthermore, there was a gender effect: women, but not men, showed a significant effect of priming on agency ratings, consistent with previous results (Jones et al., 2008). As expected, checking symptoms were found to be negatively related to experienced agency in primed trials; this effect was found to be specific to checking as compared to the other OCD dimensions, and once comorbid depression and gender were controlled for. In addition, checking symptoms appeared to be negatively associated with lower
542
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
experienced agency in the no-priming condition; while ordering and hoarding symptoms were found to be related to higher agency ratings in no-prime trials. Finally, there was no relationship between checking symptoms and the priming sensitivity index, although ordering symptoms tended to be negatively associated with the overall benefit of priming, but not when corrected for multiple comparisons (i.e., at an uncorrected significance). When we compared participants with high levels of checking symptoms to those without checking symptoms, there was a significant main effect of checking symptoms on agency ratings, suggesting that checking participants had overall undermined agency ratings across the two conditions. Further analyses revealed no differences between the two groups in terms of participants’ judgment of self-causation when the priming of outcomes was absent (i.e., no-priming condition); however, checking participants scored lower than non-checking participants in the priming condition. Also, there was a trend towards significance for the interaction between checking and the priming effect, suggesting that priming did not enhance experienced agency in high-checking participants. The finding that individuals with checking symptoms were less prone to experience agency for consistent action-effects supports the assumption that checking phenomena are characterized by an undermined ‘‘feeling of doing.” Considering that, in everyday behaviors, the feeling of doing may constitute an important cue for goal satisfaction and subsequent action closure (Aarts et al., 2005; Szechtman & Woody, 2004; Woody & Szechtman, 2000; Woody et al., 2005), an attenuation of this kind of phenomenal cue could explain why checking individuals frequently experience incompleteness, ‘‘not quite right” feelings and doubts about goal achievement. From this perspective, checking symptoms constitute behavioral adjustments in order to get more convincing (explicit) cues about actual goal completion or to experience a complete feeling of doing (i.e., ‘‘just right feelings”). Furthermore, Aarts et al. (2005) suggested that experienced agency elicited in the illusion of agency paradigm may depend on an unconscious mechanism of comparison that triggers a matching signal whenever a correspondence between observed and anticipated effects (i.e., effect-priming prior to action) is perceived. Along the same lines, Pitman (1987) proposed that recurrent inconsistent mismatch signals may underlie repeated behaviors in OCD. Thus, our results support the hypothesized involvement of a dysfunctional comparing mechanism and related matching signals in checking symptoms, as well as the idea of an attenuated subjective experience in connection with action-monitoring difficulties (Szechtman & Woody, 2004; Woody et al., 2005). Consistent with our hypothesis, altered action awareness may be specific to checking and not the other OCD dimensions. Indeed, as expected, the other OCD dimensions did not share any specific relationship with the measures of experienced agency; ordering symptoms were found to be negatively associated with the overall benefit of priming, but not when corrected for multiple comparisons (i.e., at an uncorrected significance). The associations found between lower experienced agency and checking and, to a lesser extent, ordering symptoms are quite consistent with the two-dimensional model of OCD whereby those symptoms may be predominantly associated with a peculiar experience of incompleteness. Nevertheless, one might expect that OCD symptoms frequently associated with the overestimation of personal influence in the occurrence of random outcomes (i.e., obsessing and washing dimensions) would be characterized by an increased experience of agency for uncontrollable outcome that match prior thoughts. One possible explanation is that the exaggerated personal responsibility that may characterize washing and obsessing symptoms may depend on the valence of the content of preceding thoughts and uncontrolled events (i.e., threatening events vs. neutral outcomes, as in the present study). It should be noted that participants with OCD symptoms (regardless of OCD subtype) have been reported to experience an increased illusory sense of control when desired outcomes match actual outcomes; this effect appeared more marked with aversive stimuli than with neutral stimuli (Reuven-Magril, Dar, & Liberman, 2008). In their study, Reuven-Magril et al. used a preprogrammed sequence of neutral and aversive images, for which participants had to control and shorten their duration (i.e., desired outcome) by finding the right combination of key presses (i.e., action); their perceived level of control was assessed at various points during the task. Participants did not have any actual control over the duration of the stimuli, as the presentation time gradually increased (i.e., increased discrepancy between desired and actual outcomes) throughout the first half of the task and then gradually decreased for the remaining stimuli (i.e., increased matching between desired and actual outcome). Their results suggest that OCD symptoms may be characterized by more compulsive-like repetitive patterns of action (i.e., using the same key presses) and by an increased sense of control for aversive and, to a lesser extent, neutral stimuli. The apparent contradiction between the results of our study and the findings of Reuven-Magril et al. (2008) may be related to some major differences between the illusion of agency task and the illusion of control task. First, the two tasks may differ regarding the kind of sense of control they assess. Indeed, a recent article on the contribution of sense of control on sense of agency proposed to distinguish between the ‘‘sense that one has to exert control to generate and maintain an appropriate action program” and the ‘‘sense that one feels in control of an action” (Pacherie, 2008, p. 20). The first form of control depends on the importance of mental and behavioral attempts and the adjustments necessary to reduce the discrepancies between expectations and outcomes (i.e., ‘‘effortful control,” such as would be the case in disrupted or unfamiliar actions); in such effortful situations, it is the conscious effort itself that may enhance the sense that one is engaged in and causing actions. On the other hand, feeling in control of an action only requires the accessibility of the result of the unconscious comparisons between predicted and actual states (‘‘effortless control,” such as would be the case for routine or automatic actions; Pacherie, 2008). In light of this model, we could argue that, in the illusion of control task used by Reuven-Magril et al., feelings of control and self-causation are predominantly elicited by the effortful control situation created by the design of the task. In our study, on the other hand, the illusion of agency task elicited a sense of effortless control and thus feelings of self-causation were predominantly based on automatic and unconscious processes, similar to those observed during nonconscious goal pursuit (Aarts et al., 2005, 2006).
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
543
Furthermore, a recent cognitive model of OCD (Moulding & Kyrios, 2006, 2007) suggests the involvement of two controlrelated features: the need for control (i.e., the general motivation to be able to exert control over events) and the sense of control (i.e., beliefs in one’s ability to attain or avoid specific outcomes through one’s actions). According to this model, the level of desired control generally leads to the deployment of appropriate behaviors in order to achieve desired outcomes; the level of perceived control in a specific context may vary accordingly. While, in most people, sense of control is generally positively related to the desire for control, a significant discrepancy between desired level of control and perceived control in certain situations (i.e., a high need for control but a low-level of perceived control) may lead OCD individuals to develop strategies that will give them a satisfying ‘‘feeling that things are under control.” Thus, it is possible that the Reuven-Magril et al. (2008) paradigm may offer a context favoring such an artificially inflated sense of control, while the task used in our study may emphasize a process that is potentially implicated in the undermined sense of control, which has been particularly related to checking symptoms (Moulding & Kyrios, 2007). In addition, the Reuven-Magril et al. study may highlight the phenomenon whereby compulsive-like behaviors lead OCD individuals to experience high levels of control, rather than the process underlying anxiety and distress, which in turn trigger OCD symptoms. In other words, the ‘‘feelings of undesired end-state to be realized” that characterize some OCD symptoms and subsequently trigger avoidance strategies remain to be explored; thus, further studies should be conducted to explore the relationship between obsessing and washing symptoms and the overestimation of personal influence in the occurrence of magical and/or threatening outcomes (Pronin, Wegner, McCarthy, & Rodriguez, 2006). Another possibility is that harm avoidance could be related to a more conceptual component of action awareness, namely the feeling that one’s conscious thought is the cause of one’s action (Pacherie, 2008; Synofzik, Vosgerau, & Newen, 2008; Wegner & Sparrow, 2004). This ‘‘conceptual sense of agency” would be based on conscious inferences leading one to draw (illogical) causal links between one’s thoughts, identified as intentions, and external outcomes (i.e., a kind of post hoc ergo propter hoc fallacy). Exaggerated responsibility for threat events, which in fact are not the result of one’s own actions but did coincide with one’s consistent thoughts, would reflect an impaired top-down processing of agency (Synofzik et al., 2008), leading to a bias towards the self in the attribution of the agency of an event. By contrast, checking symptoms would be related to a ‘‘phenomenal sense of agency,” implying the feeling that one’s movement is causing some specific effects. This component may depend on an unconscious matching signal between actual outcomes of movement and anticipated effects, which may stem from bottom-up processes (Synofzik et al., 2008). However, these assumptions are merely speculative and laboratory experiments are needed to explore the extent to which the sense of agency constitutes a common factor in OCD dimensions, with the possibility that impairment of specific components of this sense could lead to different patterns of OCD symptoms. Before concluding, we should emphasize some limitations of the present study. First, our interpretation of the nature of the relationship found between experienced agency and checking symptoms is limited, as we did not control for the potential confounding effect of other factors, such as dissociation, anxiety, and attentional focus. For example, dissociative states, which are frequently associated with checking (Belayachi & Van der Linden, 2009; Rufer, Fricke, Held, Cremer, & Hand, 2006), can reduce awareness of action by altering the integration of action representations (e.g., expected effects) and the ‘‘present self in action” (e.g., observed outcomes) (Ansfield & Wegner, 1996). Furthermore, several studies have emphasized the importance of attention focused on representations related to goals, expected effects or intention in action awareness, as compared to attention focused on movements (Lau, Rogers, Haggard, & Passingham, 2004). Consistently with this idea, checking individuals seem to parse their habitual actions mainly with regard to movement parameters, rather than goals and outcomes (Belayachi & Van der Linden, 2009). However, anxiety, which focuses attention on details and promotes local perceptual information processing, could alter the shift of attention towards effect representation (Boyer & Liénard, 2006; Derryberry & Reed, 1998). Thus, as we did not control for the impact of anxiety, attention and dissociation in this study, we cannot rule out the possibility that those factors may account, at least in part, for the relationship found between checking symptoms and low sense of agency. Further studies should therefore explore the potential confounding effect of those variables. Furthermore, our results prevent us from determining the nature of the relationship between lower experienced agency and compulsive checking. Indeed, what remains unclear is whether lower experience of agency leads people to check their actions in order to ensure that they have been completed, or whether repeated checking of actions may lead people to deploy alternative strategies for authorship ascription. It is possible that a lack of matching signal accessibility may prevent people from experiencing success in achieving expected outcomes, which may in turn lead them to check their actions. Conversely, individuals with checking symptoms may rely on different cues for authorship processing, such as movement parameters. One way of answering this question would be to explore the extent to which priming the concept of success may increase feelings of self-causation in checking individuals, and the extent to which inducing checking may engender an undermined experienced agency and less confidence in self-agency. Nevertheless, we assumed that an impaired feeling of doing may constitute only one of the numerous factors implicated in checking, such as anxiety, attentional focus, reality monitoring difficulties and poor memory for actions (Boyer & Liénard, 2006; Zermatten & Van der Linden, 2008b). As mentioned above, anxiety and attentional focus may have a potential confounding effect as those variables may interfere with the normal processing of action-monitoring. Furthermore, an undermined immediate sense of doing could be responsible for reality monitoring difficulties and memory failure, which in turn may trigger doubts about action performance and checking symptoms. In this vein, it would be important to clarify the relationships between low feeling of doing, difficulties with reality monitoring and action memory.
544
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
Another potential limitation is that the study also lacks a control for the impact of beliefs about self-agency on agency ratings. Indeed, checking symptoms correlated with lower experienced agency in both primed and non-primed trials; group comparisons consistently revealed a main effect of checking proneness on overall agency ratings, suggesting that checking participants reported lower agency ratings than those with no checking symptoms. In light of those results, one could argue that checking individuals have lower agency ratings in both the primed and non-primed conditions because they have an overall lack of confidence in their self-agency. Thus, an alternative explanation of our results would be that the relationship observed between authorship ascription and checking symptoms could be mediated by control beliefs such as self-determination, which have recently been found to constitute an additional source of information for authorship processing (Aarts, Oikawa, & Oikawa, 2010). Consistently with this idea, OCD symptoms, and particularly checking symptoms, may be related to an undermined perceived control (beliefs about one’s ability to perform an action and the extent to which that action will lead to a desired outcome or avoid an undesirable one; Moulding & Kyrios, 2007). Thus, further studies are needed to explore the extent to which beliefs of control and self-determination may account, at least in part, for the relationship found between checking symptoms and lower self-agency. Nonetheless, group comparisons also revealed that checking participants scored significantly lower than non-checking participants on primed trials, while there was no group difference on non-primed trials, supporting the assumption that checking participants were less prone to base their agency on expected outcomes. However, there are limits to how to interpret unexpected results concerning agency ratings in the no priming of outcome condition as there is no assumption about the mechanisms that underlie agency judgments in the absence of ‘‘representation of action-effect”(i.e., the observed outcome was not primed prior action). The action control mechanism underlying agency ratings in the Wheel of Fortune Task may vary across the two conditions (e.g., conceptual congruency between preview and action-effect in the priming condition vs. congruency between sensory predictions made from the movement and the actual sensory consequence in the no-priming condition). Therefore further examinations of cues that might allow participants to correctly misidentify the self as the cause of observed outcomes in the no-priming condition would help to clarify some of our results. Another intriguing outcome is that ordering symptoms were negatively correlated with priming sensitivity, whereas checking symptoms were not, although group comparisons revealed an absence of any effect of priming for the checking participants, as compared to non-checking participants. However, it is worth noticing that the negative association found between ordering symptoms and the priming sensitivity index is difficult to interpret. Indeed, this agency measure, which has been defined as the mean agency ratings in primed trials less mean agency ratings on unprimed trials, may range from 9 to 9. Thus, negative scores suggest a negative effect of priming (i.e., resulting in lower agency ratings on primed trials than on non-primed trials), positive scores suggest a benefit of priming (i.e., resulting in higher agency ratings on primed trials than on non-primed trials), while a null score suggests an absence of any priming effect. However, we cannot determine the extent to which a negative correlation with this measure reflects a negative effect of priming on agency ratings, an absence of priming effect or a lower priming effect (i.e., a benefit of priming, but to a lesser extent). Thus, the results obtained from the priming sensitivity index and the related interpretations must be considered cautiously. In addition, the OCI-R alone overlooks the motivational aspects of checking symptoms, whereas some studies have yielded evidence that checking symptoms are motivationally heterogeneous (e.g., Ecker & Gönner, 2008; Tolin, Woods, & Abramowitz, 2003). Thus, what remains unclear is whether checking symptoms, as measured by the OCI-R checking subscale, reflect a tendency to re-enact actions because of a lack of an immediate feeling of satisfaction (incompleteness) or compulsive checking of safety actions in order to avoid potential threat (harm avoidance). Also, it is possible that incompleteness feelings may trigger doubts about whether an action has been achieved, which in turn can lead to harm avoidance features whenever one becomes aware of the danger, if such doubts concern safety actions, and consequently lead to checking behaviors (i.e., incompleteness and harm avoidance). Such heterogeneity within the checking measurement may possibly account for the weak association we observed between checking symptoms and experienced agency. Further studies should be conducted in order to compare various subgroups of checkers according to their predominant motivation, as well as OCD subtypes within the same core motivational dimension, on the illusion of agency task. Finally, another important limitation of this study is the relative weakness of the effect of checking proneness on undermined self-agency. In this regard, we should emphasize that our study does not control for major factors that potentially interfere with the relationship between checking symptoms and self-agency as elicited by the Wheel of Fortune Task, such as the way participants generally construe their actions (i.e., level of agency). Indeed, this factor has been related to both checking symptoms (Belayachi & Van der Linden, 2009) and self-agency based on the Wheel of Fortune Task (van der Weiden, Aarts, & Ruys, 2010). Level of agency stems from the Action Identification theory (Vallacher & Wegner, 1985, 1989), which states that any action can be identified within a cognitive hierarchy of meanings, in which the lower-levels represent instrumental features and movement parameters, while the higher-levels relate to the desired goal and expected outcomes. Vallacher and Wegner proposed that the particular level at which an action is identified may reflect the particular representation (movements parameters vs. outcome) on the basis of which the action is monitored. Although the way an action is identified depends on several action-related features (e.g., action complexity, degree of expertise, action disruption or error), people tend to adopt a predominant level of action identification for various behaviors (i.e., level of agency; Vallacher & Wegner, 1989). Thus, there are people who have a tendency to identify their actions according to related purpose and implications (i.e., a high-level of agency) and others who tend to identify their actions mainly regarding procedural aspects and motor subcomponents (i.e., a low-level of agency). Recently, van der Weiden et al. examined the relationship between the level of agency and the feeling of doing by using the Behavior Identification Form, a questionnaire designed to assess
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
545
the level at which participants generally construe their habitual actions (i.e., mainly based on movement parameters vs. expected outcomes), in addition to the Wheel of Fortune Task. Their results suggested that the more people tend to generally identify their action according to related outcomes and pursued goals (i.e., high-level of agency), the more they experience self-agency for outcomes that have been primed prior actions. However, outcome priming did not increase agency ratings among participants who had a low-level of agency. Interestingly, checking symptoms have been previously found to be related to a low-level of agency, as assessed by the Behavior Identification Form (Belayachi & Van der Linden, 2009). Furthermore, recent data (Belayachi & Van der Linden, 2010), suggest that checking participants with no dysfunctional beliefs are those who are also characterized by a low-level of agency. Taken as a whole, one could argue that the rather small effect sizes in our study could be related to the lack of control of both the level of action construal and the motivational core feature of checking symptoms. Further studies are needed in order to clarify this issue. In sum, our results suggest that individuals with checking symptoms have an impaired ability to experience agency for outcomes that match expectations, while the other OCD dimensions are not related to impaired agency. These results also suggest that an impaired sense of agency in checking symptoms is likely to be connected with an alteration of basic action processing involved in the phenomenal sense of acting and causing effects congruent to ‘‘what was intended.” Overall, they are in agreement with the idea that abnormal action-monitoring, observed in both clinical and non-clinical samples, may underlie checking features. Nevertheless, although studies of non-clinical samples can generally be extrapolated to clinical samples, further studies should be conducted to replicate our findings in individuals with more severe OCD symptoms (i.e., OCD patients). In addition, further experiments are needed to explore the extent to which some OCD symptoms may be specifically related to another component of sense of agency potentially involved in altered personal causation in the occurrence of uncontrollable events. Thus, it would be interesting to explore whether the two core motivational dimensions in OCD may be at least partly explained by the component of the sense of agency that is predominantly impaired. Acknowledgment This study was supported by a Grant from the French-speaking community of Belgium (Action Recherche Concertée, Convention 06/11-340). References Aarts, H., Custers, R., & Wegner, D. M. (2005). On the inference of personal authorship: Enhancing experienced agency by priming effect information. Consciousness and Cognition, 14, 439–458. Aarts, H., Oikawa, M., & Oikawa, H. (2010). Cultural and universal routes to authorship ascription: Effects of outcome priming on experienced self-agency in the Netherlands and Japan. Journal of Cross-Cultural Psychology, 41, 87–98. Aarts, H., Wegner, D. M., & Dijksterhuis, A. (2006). On the feeling of doing: Dysphoria and the implicit modulation of authorship ascription. Behaviour Research and Therapy, 44, 1621–1627. Allison, P. D. (1999). Logistic regression using the SAS system: Theory and application. Cary, NC: SAS Institute Inc. Ansfield, M. E., & Wegner, D. M. (1996). The feeling of doing. In P. M. Gollwitzer & J. A. Bargh (Eds.), The psychology of action: Linking cognition and motivation to behavior (pp. 482–506). New York: Guilford Press. Belayachi, S., Van der Linden, M. (2010). Level of agency in subclinical checking: A dysfunctional beliefs-based effect, in preparation. Belayachi, S., & Van der Linden, M. (2009). Level of agency in sub-clinical checking. Consciousness and Cognition, 18, 293–299. Boyer, P., & Liénard, P. (2006). Why ritualized behavior? Precaution systems and action parsing in developmental, pathological and cultural rituals. Behavioral and Brain Sciences, 29, 1–56. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Coles, M. E., Frost, R. O., Heimberg, R. G., & Rhéaume, J. (2003). ‘‘Not just right experiences”: Perfectionism, obsessive–compulsive features and general psychopathology. Behaviour Research and Therapy, 41, 681–700. Derryberry, D., & Reed, M. A. (1998). Anxiety and attentional focusing: Trait, state and hemispheric influences. Personality and Individual Differences, 25, 745–761. Ecker, W., & Gönner, S. (2008). Incompleteness and harm avoidance in OCD symptom dimensions. Behaviour Research and Therapy, 46, 895–904. Fitzgerald, K. D., Welsh, R. C., Gehring, W. J., Abelson, J. L., Himle, J. A., Liberzon, I., et al (2005). Error-related hyperactivity of the anterior cingulate cortex in obsessive–compulsive disorder. Biological Psychiatry, 57, 287–294. Foa, E. B., Huppert, J. D., Leiberg, S., Langner, R., Kichic, R., Hajcak, G., et al (2002). The Obsessive–Compulsive Inventory: Development and validation of a short version. Psychological Assessment, 14, 485–496. Fox, J. (1991). Regression diagnostics. Thousand Oaks, CA: Sage Publications. Fuhrer, R., & Rouillon, F. (1989). La version française de l’échelle CES-D (Center for Epidemiologic Studies – Depression Scale). Description et traduction de l’échelle d’auto-évaluation. Psychiatrie et Psychobiologie, 4, 163–166. Gehring, W. J., Himle, J., & Nisenson, L. G. (2000). Action monitoring dysfunction in obsessive–compulsive disorder. Psychological Science, 11, 1–6. Hajcak, G., & Simons, R. F. (2002). Error-related brain activity in obsessive–compulsive undergraduates. Psychiatry Research, 110, 63–72. Jones, S. R., de-Wit, L., Fernyhough, C., & Meins, E. (2008). A new spin on the Wheel of Fortune: Priming of action-authorship judgements and relation to psychosis-like experiences. Consciousness and Cognition, 17, 576–586. Lau, H. C., Rogers, R. D., Haggard, P., & Passingham, R. E. (2004). Attention to intention. Science, 303, 1208–1210. Maltby, N., Tolin, D. F., Worhunsky, P., O’Keefe, T. M., & Kiehl, K. A. (2005). Dysfunctional action monitoring hyperactivates frontalstriatal circuits in obsessive–compulsive disorder: an event-related fMRI study. Neuroimage, 24, 495–503. Moulding, R., & Kyrios, M. (2006). Anxiety disorders and control related beliefs: The exemplar of obsessive–compulsive disorder (OCD). Clinical Psychology Review, 26, 573–583. Moulding, R., & Kyrios, M. (2007). Desire for control, sense of control and obsessive–compulsive symptoms. Cognitive Therapy and Research, 31, 759–772. Muris, P., Merckelbach, H., & Clavan, M. (1997). Abnormal and normal compulsions. Behaviour Research and Therapy, 35, 249–252. Pacherie, E. (2008). The phenomenology of action: A conceptual framework. Cognition, 107, 179–217. Pitman, R. K. (1987). A cybernetic model of obsessive–compulsive pathology. Comprehensive Psychiatry, 28, 334–343. Pronin, E., Wegner, D. M., McCarthy, K., & Rodriguez, S. (2006). Everyday magical powers: The role of apparent mental causation in the overestimation of personal influence. Journal of Personality and Social Psychology, 91, 218–231. Rachman, S. (1997). A cognitive theory of obsessions. Behaviour Research and Therapy, 35, 793–802.
546
S. Belayachi, M. Van der Linden / Consciousness and Cognition 19 (2010) 534–546
Radloff, L. S. (1977). The CES-D Scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385–401. Reuven-Magril, O., Dar, R., & Liberman, N. (2008). Illusion of control and behavioral control attempts in Obsessive–Compulsive Disorder. Journal of Abnormal Psychology, 117, 334–341. Ridderinkhof, K. R., van den Wildenberg, W. P. M., Segalowitz, S. J., & Carter, C. S. (2004). Neurocognitive mechanisms of cognitive control: The role of prefrontal cortex in action selection, response inhibition, performance monitoring, and reward-based learning. Brain and Cognition, 56, 129–140. Riddle, A. S., Blais, M. R., Hess, U. (2002). A multi-group investigation of the CES-D’s measurement structure across adolescents, young adults and middle-aged adults, CIRANO, Montreal (QC) (2002) Working Paper No. 2002s-36. Rufer, M., Fricke, S., Held, D., Cremer, J., & Hand, I. (2006). Dissociation and symptom dimensions of obsessive–compulsive disorder: A replication study. European Archives of Psychiatry and Clinical Neuroscience, 256, 146–150. Salkovskis, P. M. (1985). Obsessional–compulsive problems: A cognitive-behavioural analysis. Behaviour Research and Therapy, 23, 571–583. Summerfeldt, L. J., Huta, V., & Swinson, R. P. (1998). Personality and obsessive–compulsive disorder. In R. P. Swinson, M. M. Antony, S. Rachman, & M. A. Richter (Eds.), Obsessive–compulsive disorder: Theory, research, and treatment (pp. 79–119). New York: Guilford Press. Synofzik, M., Vosgerau, G., & Newen, A. (2008). Beyond the comparator model: A multifactorial two-step account of agency. Consciousness and Cognition, 17, 219–239. Szechtman, H., & Woody, E. Z. (2004). Obsessive–compulsive disorder as a disturbance of security motivation. Psychological Review, 111, 111–127. Tolin, D. F., Brady, R. E., & Hannan, S. (2008). Obsessional beliefs and symptoms of obsessive–compulsive disorder in a clinical sample. Journal of Psychopathology and Behavioral Assessment, 30, 31–42. Tolin, D. F., Woods, C. M., & Abramowitz, J. S. (2003). Relationship between obsessive beliefs and obsessive–compulsive symptoms. Cognitive Therapy and Research, 27, 657–669. Ursu, S., Stenger, V. A., Shear, M. K., Jones, M. R., & Carter, C. S. (2003). Overactive action monitoring in obsessive–compulsive disorder. Psychological Science, 14, 347–353. Vallacher, R. R., & Wegner, D. M. (1985). A theory of action identification. Hillsdale, NJ: Lawrence Erlbaum. Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual variation in action identification. Journal of Personality and Social Psychology, 57, 660–671. van der Weiden, A., Aarts, H., & Ruys, K. (2010). Reflecting on the action or its outcome: Behavior representation level modulates high level outcome priming effects on self-agency experiences. Consciousness and Cognition. Wegner, D. M. (2002). The illusion of conscious will. Cambridge, MA: MIT Press. Wegner, D. M., & Sparrow, B. (2004). Authorship processing. In M. Gazzaniga (Ed.), The new cognitive neurosciences (3rd ed., pp. 1201–1209). Cambridge, MA: MIT Press. Wegner, D. M., & Wheatley, T. P. (1999). Apparent mental causation: Sources of the experience of will. American Psychologist, 54, 480–492. Woody, E. Z., Lewis, V., Snider, L., Grant, H., Kamath, M., & Szechtman, H. (2005). Induction of compulsive-like washing by blocking the feeling of knowing: An experimental test of the security-motivation hypothesis of obsessive–compulsive disorder. Behavioral and Brain Functions, 1, 1–11. Woody, E., & Szechtman, H. (2000). Hypnotic hallucinations and yedasentience. Contemporary Hypnosis, 17, 26–31. Zermatten, A., & Van der Linden, M. (2008a). Impulsivity in non-clinical persons with obsessive–compulsive symptoms. Personality and Individual Differences, 44, 1824–1830. Zermatten, A., & Van der Linden, M. (2008b). Phenomenal characteristics of memories of daily actions in checking-prone individuals. Applied Cognitive Psychology, 22, 1099–1112. Zermatten, A., Van der Linden, M., Jermann, F., & Ceschi, G. (2006). Validation of a French version of the Obsessive–Compulsive Inventory – Revised in a nonclinical sample. European Review of Applied Psychology, 56, 151–208. Zermatten, A., Van der Linden, M., Larøi, F., & Ceschi, G. (2006). Reality monitoring and motor memory in checking-prone individuals. Journal of Anxiety Disorders, 20, 580–596.
Consciousness and Cognition 19 (2010) 547–579
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
The Endogenous Feedback Network: A new approach to the comprehensive study of consciousness Claudia Carrara Augustenborg University of Copenhagen, Institute for Psychology, Oester Farimagsgade 2A, 1353 København K, Denmark
a r t i c l e
i n f o
Article history: Received 17 September 2009 Available online 3 April 2010 Keywords: Neural networks Endogenous feedback Awareness Self-consciousness Self-awareness Non-cognitive consciousness Phenomenal consciousness Access-consciousness Monitoring consciousness
a b s t r a c t The phenomenon of consciousness has received through the centuries a profusion of interpretations, engaging researchers across many disciplines. Nevertheless, consensus still floats at large. The aim of the present work is therefore twofold. Through the review of a selected number of existing proposals, it will first be considered the extents of their reciprocal compatibility, tentatively shaping an integrated, theoretical profile of consciousness. A new theory, the Endogenous Feedback Network (EFN) will consequently be introduced which, besides being able to accommodate the main tenets of the reviewed theories, appears able to compensate for the explanatory gaps they leave behind. The EFN proposes consciousness as the phenomenon emerging from a distinct neural network broadcasting the changes associated to mental activations across the brain. It additionally argues for the need to include a 5th element to Ned Block’s taxonomy (i.e. phenomenal, monitoring, access, and self-consciousness) that is, non-cognitive consciousness. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Although consciousness has been a topic of interest along centuries and across disciplines, it has been in the last decades that theories of consciousness – both philosophical and neuroscientific1 – have flourished with particular vigor. A probable reason for the modern luxuriance of the theoretical landscape might lie in the technological progresses that in the late years have armed the hands of scientific investigation. In fact, as brain electrophysiological recording and imaging techniques allow researchers to enter with increasing accuracy the intricate neural labyrinths of our brains, so are nonviable philosophical tenets crossed over, spurring the shaping of new hypotheses. Nevertheless, although many proposals present brilliant intuitions, and offer seemingly valid explanations for the variety of mechanisms connected to the emergence and unfolding of consciousness, none of them is yet able to fully account for the phenomenon as a whole. As Delacour (1997) pointedly argued, the real obstacle toward the neurobiology of consciousness does not reside as much in the methodological limits intrinsic in the attempt to transform the subjectivity of human experience in an objective topic of research, as rather in the theoretical framing of the phenomenon. Mischievously, consciousness seems in fact to expect from its investigators first of all the ability to carve a coherent path in the confusing mist of different and often contradictory meanings given to it and to its many aspects. Terms like awareness, conscious states, self-consciousness, self-awareness and phenomenal experience can assume quite different meanings depending on the context, and on the academical background of the investigator. Then again, it is not rare to find these terms used as reciprocal synonymous in the pages of a same article.
E-mail address:
[email protected] Wishing to accommodate within a wide disciplinary umbrella a variegated number of perspectives and research, it will be adopted the comprehensive term ‘neuroscience’ throughout the paper, also when referring to work that, in different contexts, would more precisely fall within the frames of neuropsychology. 1
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.007
548
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Considering that the search for consciousness will then take the lead from the specific conceptualization chosen to address the phenomenon, and at the light of the persisting terminological intricacy, it should therefore be of little surprise that consensus within and between disciplines has yet to be reached. While uniformity of thought is generally not an auspicious feature of research, it might be equally true that constructive efforts toward given degrees of theoretical agreement can indeed contribute to strengthen the values of the respective empirical achievements. In other words, before being able to determine whether or not we have found the object of our search, we should agree about what it is we are actually looking for. The first question to ask should therefore be, ‘‘What does neuroscience mean by consciousness?”. Some theories might suggest that consciousness is wakefulness as opposite to sleep and coma (as it seems implicit in the practice to measure consciousness on the basis of the subject’s responsiveness; see Giacino & Malone, 2008). However, different accounts could argue that consciousness is the awareness of our mental states, and consciousness and wakefulness are not necessarily related (see Bosinelli, 1995). A third approach might then propose consciousness as synonymous with phenomenal experience, since it conveys the feeling of our mental acts (see Damasio, 1999; Thagard & Aubie, 2008). Each of these assumptions would then determine a given research path, which might eventually produce results supporting the specific approach, but that nevertheless might fail to impress the bearers of different theoretical tenets. However, such apparent scientific disagreements might depend – at least in the theoretical scenarios offered above – on the implicit assumption that terms like ‘consciousness’, ‘awareness’, and ‘experience’ should refer to a same phenomenon. On the other hand, it could be that such varied terminology might truly describe different aspects of the same phenomenon. The second question we might want to ask could be, ‘‘How does consciousness emerge?”. Atkinson, Thomas, and Cleeremans (2000) suggested two possible, initial sources: Consciousness could either be the product of special computations, or the product of some types of representations. Either one of these options would then ask for a further choice: Consciousness could emerge from a structure dedicated to it, or consciousness could be the result of some special computation occurring anywhere in the brain. The combinatorial possibilities emerging from such variety of approaches will then predictively lead to equal richness of hypotheses and of experimental designs. However, by shaping a conceptual distinction between different aspects of consciousness, it could also be the case that different approaches could quite accurately be adopted to describe different stages of the phenomenon. Finally, we could attempt a third question, ‘‘What do we need consciousness for?”. The question gains more interest if we consider the evidence showing that a great deal of our actions is carried on without us being consciously aware of it (Wegner, 2002), and that often our performances improve if under the control of the automatic system2 (Bargh, 2002). On the other hand, DeWall and colleagues (2008) argued that consciousness is indeed vital for our logical reasoning, since it is responsible for the integration and manipulation of large amounts of stimuli. Choosing a different approach, Rosenthal (2008) argued that consciousness has not emerged as a benefit per se, rather as a by-product of other adaptive features of human cognition, e.g. language (having developed the ability to talk and to communicate externally, we received the bonus of being able to turn our voices inwardly). To the contrary, Bering and Shackelford (2004) have proposed that consciousness might not be a product of evolution, rather the promoter of it. Nevertheless, given the adoption of clear and unambiguous terminology, and a rigorous conceptual demarcation between the various aspects of consciousness, a significative degree of agreement about the role of consciousness might emerge even between such varied row of proposals. Taking the step from the assumption that no theory – whether it belongs to psychology or to philosophy – is likely completely mistaking in its account of consciousness, this article’s initial intent will in fact be to interweave a theoretical connective tissue that could unify at least some of the knowledge accumulated through decades of neuroscientific efforts, and centuries of philosophical speculations around the mystery of consciousness. The way such goal will be pursued will first of all consist in trying to merge – intra- and inter-disciplinarily – the features of reciprocal compatibility between a selected group of philosophical accounts (see Table 1), and the points of strength of some among the existing neuroscientific theories of consciousness (see Section 1). Adopting a strictly neuroscientific approach to the phenomenon, this article will then introduce a new theory of consciousness – Endogenous Feedback Network theory (EFN) – which respects the positive accomplishments of all reviewed theories, embracing them under a same theoretical umbrella, and consequently accommodating much of their respective evidence and tenets. Even more ambitiously, the new theory of consciousness further tries to fill in the gaps within and between the existing theories by offering slight modifications and/or appropriate theoretical addenda. The introduction of non-cognitive consciousness as a 5th element to be integrated to Block’s taxonomy of the phenomenon represents one of such theoretical additions. Along with its specific conceptualization of consciousness, the EFN theory will also ascribe distinct meanings to the terminology surrounding the phenomenon (Box 1). The present work will conclude with a general appraisal of the strengths and limits of the Endogenous Feedback Network theory, and with the evaluation of its degree of compatibility with the theories of consciousness reviewed ahead.
2 The argument is based on the assumption that, in order to make our choices fit our conscious desires, we tend to build ‘pseudo-rational’ confabulations that might justify specific decisions, even when these might objectively not represent the best options. Decisions taken unconsciously are supposed to be immune from such distortions.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
549
2. Section 1 2.1. Neuroscience of consciousness Standing the massive and variegated bulk of literature, this section clearly does not hold the presumptuousness of being able to offer an exhaustive account of such manifold of neuroscientific approaches to consciousness. Only a selected group of theories will consequently appear in this section, penalty being a reduced account on advances and limits of our knowledge on the topic, and a more limited perspective on the nature of the phenomenon. However, such selection has not been operated on the basis of theories assumed validity, but rather of their being representative of the specific approaches. Nevertheless, although each pivoting on their respective premises, these theories are all bound by some common assumptions: Consciousness is a state of the brain, it has a material reality and it has a causal power (Delacour, 1997). 2.1.1. Ned Block: Mapping consciousness Ned Block has not attempted to propose an overall account of consciousness. Nevertheless, his contribution covers a role of great relevance in setting the frames for the multiple features of the phenomenon. For this reason, it seems appropriate to let a summarization of his work paving the road to the theories of consciousness that will be presented in the remaining of this section. Block (1995) considered consciousness comprising of four different aspects, access-, monitoring-, phenomenal and selfconsciousness, and he consequently suggested that we should look for different neural correlates of consciousness (NCCs), each determining a different feature of consciousness. In his terms, we should speak of core NCC as the part that shapes the specific aspect of consciousness, and total NCC as the other neural element that makes the core able to emerge. Hereby it will be tried to sketch Block’s definitions of the above-mentioned features of consciousness. Access-consciousness refers to the ability of an organism to acquire awareness about own mental states, and to elaborate upon the information processed by own cognitive mechanisms. Borrowing Braisby’s (2002) terminology, access-consciousness can be defined as a reflexive analysis upon own reasoning. Block considered access-consciousness a cluster concept in which the content of the representation emerges in reasoning, planning of actions and speech (Block, 1995). However, with reference to the latter function, Block considered it only marginally relevant: In fact, reportability is for Block not a compulsory feature of access-consciousness, as it is the mental presence of the inner representation that matters. By this token, it is assumed possible that also non-linguistic animals could experience access-consciousness. Defined by Block as the ability to maintain an internal scanning of oneself, monitoring consciousness is the feature of consciousness which supposedly supervises very familiar action patterns. In fact, although being highly practiced and therefore in no need of great cognitive resources, common behavioral sequences still require supervision in order not to interfere with similar ones (Block, 1995). Already in 1979, Reason (in Cohen (2003)) suspected that the failure of some specific kind of consciousness could have been responsible for the faulty triggering of action patterns that would consequently overshadow different original goals. Self-consciousness addresses then the individual’s awareness of the self, both in terms of own mental states and perceptions, and of the knowledge of being the subject of an autobiographical narrative. Block suggested that self-consciousness is present in a number of higher primates, and it is manifested through their ability to recognize themselves in mirrors. As dogs, monkeys and quite a large variety of other animals treat themselves as strangers when their own image reflects in a mirror, their ability to experience self-consciousness could be doubted. The final aspect of consciousness, according to the suggested taxonomy, is phenomenal consciousness, which refers to experiential states. Nagel (1974) described it as the feature of consciousness that communicates the ‘‘what-is-likeness” of an experience. According to Block, phenomenal consciousness includes the experiential properties of feelings and perceptions (pain, smells, etc.), as well as of thoughts, wishes and emotions (Block, 1995). He suggested that the subjective flavor of the experience could be an intrinsic property of its content, but he also admitted that the presence of phenomenal consciousness is highly controversial. Its challenge resides in fact in the difficulty to explain the nature of a representation which, although having been deprived of all references to perceptual details such as temperature, form, or color, can still generate an impression. Resting on the inter-disciplinary edges between philosophy and psychology, Ned Block’s differentiation between the various aspects of consciousness carries a number of advantages. First of all, it is a useful instrument to help clearing the theoretical fog condensed above the conceptualization of consciousness. In fact, it can be of great relevance for intra- and interdisciplinary debates to agree whether a theory refers to the organism’s ability to possess awareness about own mental states, or to perceive the subjective feelings of the experience. Furthermore, on the basis of Block’s taxonomy, it is also possible to gain an evaluative parameter for the various theories of consciousness. A theory able to account for only one aspect of consciousness, while admittedly not able to satisfy entirely, could still be compatible with other theories able perhaps to illuminate different features of the phenomenon. Finally, with its concessions to animal consciousness, Block’s work enlarges the narrow methodological possibilities otherwise offered by the criterion of reportability.3 3 Many authors (e.g. Ericsson & Simon, 1993) considered verbal reports the necessary window through which we could spot consciousness. Furthermore, the very emergence of conscious states has by some (e.g. Sapir and Lee Whorf) been associated to the organism’s ability to verbally (referred to both internal and external language) represent own mental acts.
550
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Fig. 1. Stimuli from respective processual areas come to compound the floor-level representation determining the content of conscious experience. This first component of the triad may then merge with the higher-order representation of itself: If the appropriate conditions are met (floor-level and higherorder representations displaying synchronous firing both in terms of rate and time), the merging will determine the occurrence of a conscious state. Note: The higher-order representation does not reflect the content of the floor-level representation (i.e. stimuli). It constitutes a self-reflecting mental act acknowledging the floor-level representation as a whole.
Admittedly, Block’s taxonomic order might not have added much to the actual search for the mechanisms that allow consciousness to emerge, nor has he offered us any insight into the possible reasons for emotional experience. Nevertheless, as Rosenthal (2005) appropriately remarked, the separate investigation of distinct aspects of consciousness might lead us to a better understanding of the phenomenon as a whole. 2.1.2. Kriegel: The cross-order integration hypothesis (COI) Kriegel’s theory contains some of the elements presented by the HOT theory (see Table 1), and it could be seen as an attempt to shape a neuroscientific fundament to the concept of hierarchical thought construction. However, as the same Kriegel (2007) cared to stress, the main difference between the 2 approaches is that the cross-order integration hypothesis is an explanatory theory, while HOT theory basically has a descriptive value. The explanatory characteristic, together with a topdown approach,4 have been considered by Kriegel the essential features enabling a theory of consciousness to offer a plausible NCC, and to explain why a specific event would result conscious while another one might not. His COI hypothesis is assumed by him to adhere to both criteria. Kriegel identified three components behind the emergence of any conscious state: Floor-level representation, higher-order representation of the floor-level representation, and the integration between both, i.e. merging. Kriegel considered that, at any given moment, a certain number of stimuli can be perceived simultaneously. All such mental events converge then in an overall phenomenological experience which allows us to perceive, at any one time, our reality as a single, global and fluent conscious experience, rather than a composite picture of many different events. Eventual changes, affecting any of the stimuli that have entered the consciousness state, will trigger an update of the overall experience. An example might help. While your eyes are scrolling these lines, you are perhaps sitting at your desk: (1) your body can feel the pressure of the chair holding you, (2) your fingers perceive the smoothness of the pen, (3) your ears can catch sounds coming from adjacent offices or from an open window, meanwhile (4) you are processing the words you are reading trying to extrapolate their overall sense. According to COI theory, your various brain’s areas have just generated five representations, that is the above 4 floor-level representations, plus the awareness of yourself as their recipient (i.e. higher-order representation of the floor-level representations). The merging between these two types of representations constitutes Kriegel’s third component, and it determines the emergence of the conscious state. Fig. 1 is an attempt to illustrate the dynamic of COI mechanisms. 4 By top-down approach Kriegel meant that a theory of consciousness should emerge from the initial assumptions about what consciousness is and why it emerges, rather than from the wish to explain how we come to experience the phenomenon in strictly neural terms and where an eventual NCC should be located.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
551
Kriegel’s ambition has been to identify the neural substrates for each of the three components of the COI theory. He suggested therefore that the floor-level representations are generated by the specialized brain’s areas involved in the processing of the specific stimulus: Auditory cortex, V1, and so on. These computations will determine the content of consciousness, but not the emergence of consciousness: In such sense, Kriegel stresses that only the 2nd and 3rd component of the triad identify the NCC. The second element of the theory is instead supposedly responsible for more sophisticated computations, since it will be dealing with the individual’s psychological states. In fact, it will be at this stage that the higher-order thought will emerge, taking as its object the presence of the floor-level representation. Kriegel suggested that the brain structure likely able to sustain this kind of activity would be the dorsolateral prefrontal cortex (dlPFC). Kriegel defended his reasons for pointing to the dlPFC by mentioning clinical studies which have reported that schizophrenic patients, showing reduced gray matter in dlPFC, also presented impaired ability to recognize own condition (Flashman et al., 2001 in Kriegel (2007)). Schmitz and Johnson (2006) could then further support Kriegel by showing that activation in dlPFC increased when subjects had to report about own psychological states compared to when they had to assume those of significant others. Finally, studies considering blindsight5 have outlined the activation in dlPFC when the subject was conscious of the stimulus, while no reaction in this area appeared during unconscious processing (Sahraie et al., 1997). Kriegel’s suggestions turned then slightly fuzzier when he tried to point to the locus of the triad’s third and last element, the one responsible for the merging between floor-representations and higher-order representations. In fact, he speculatively linked it to the already quite controversial neural mechanisms in charge of mental binding6 in general. Accordingly, Kriegel suggested that the final component of the COI hypothesis might not translate into an anatomical structure, but it might rather be identified with the special kind of relationship created by the binding mechanism between the two kinds of representations. Such relationship could be determined by the synchronization of activation, in terms of time and rate, between different brain’s processing areas and dlPFC. To borrow Kriegel’s words, ‘‘the mechanisms that brings about this alignment is the binding mechanism, and the alignment is the binding” (Kriegel, 2007, p. 907). Having let Kriegel present his case, it can now be attempted an evaluation of his COI hypothesis which, although certainly presenting some attractive features, is nevertheless vulnerable to a number of criticisms. The first comment will refer to Kriegel’s assertion that his theory adheres to the top-down approach, which he had strongly advocated for any theory of consciousness. What he meant by top-down approach is the type of research that attempts first to define consciousness at its macro-levels, by drawing a broad map of its structure and function(s), and only at a subsequent point tries to identify its concrete mechanisms at micro-analytical levels. It might although seem like Kriegel’s theory has not soundly fulfilled such achievement. His suggestions about the dlPFC being the anatomical host of the second component of the COI’s triad (i.e. higher-order representation of the floor-level representation), appears largely deduced from the results of empirical studies, rather than guided by a specific theoretical framework.7 The flaw of possibly having confused correlation with causality can therefore not be excluded. Furthermore, the very hypothesis that only one, single area in our brains (dlPFC) is empowered with the faculty of mediating the ascent of a stimulus from floor-base level to full-fledged consciousness, admittedly leaves a trace of skepticism behind. Kriegel (2007) presented some hypothetical scenarios where damage in V4, V2 or A1 would determine the subject’s inability to obtain conscious awareness respectively of color, shape or sound of a given stimulus. However, Kriegel left unconsidered the consequences of an eventual damage in dlPFC. Presumably, this occurrence should disable the subject from the faculty of experiencing all kinds of conscious states, despite the nature of the stimuli. Since circumstances as such (e.g. coma, vegetative state8) are not univocally determined by dlPFC damage, and considering that focalized lesions in this area do not necessarily determine coma, the doubt of witnessing a correlation between dlPFC and consciousness gains new strength. A further perplexity concerns the implication of merging in a single state of consciousness a bundle of simultaneously perceived stimuli. As Kriegel proposed, in the instance that one of the involved stimuli should change, the modification will determine the emergence of an entirely new conscious state. This suggestion appears counter-intuitive, not to mention mentally burdensome. Let’s get back to the above example of sitting at the desk. In the moment a sound peaks above all others, your attention would likely polarize toward that specific stimulus: It might now be difficult to see your gain if your brain actually had to perform a rapid re-update of your overall perceptions (i.e. the chair you sit on, the pen between your fingers, and so on). In such sense, a system able to simultaneously present all the stimuli occurring at any one moment, while at the same time maintaining them separated one from another, might appear of greater adaptive value. It would in fact carry the advantage of allowing the shading of less relevant stimuli, while our full attendance could focus on a single one. Finally, it should be addresses the most cogent question to which Kriegel has not attempted to answer. The COI hypothesis suggested that a stimulus acquires conscious state if its floor-level representation occurs in synchrony (both in terms of 5 Blindsight is the neurological condition characterized by perceptual blindness following damage in V1. Although the patient is not able to report a stimulus, s/he can – above chance level – ‘guess’ its location, and offer some rough details about its appearance (see Weiskrantz (1996) for a detailed account of the disorder). 6 The mechanism through which various features of a same object (e.g. its shape, size, colour, movement, etc.) supposedly merge to the extent that we perceive them as elements of the same object-unity (see Revonsuo & Newman, 1999). 7 He eventually said that prefrontal cortex is the ‘natural’ candidate for computations of such sophistication (Kriegel, 2007, p. 903). However, as it could assumedly even be the case that no great sophistication might be needed for consciousness to emerge, it could be argued that his choice of brain structure might have been biased. 8 The absence of consciousness in vegetative state is however controversial. The given degrees of activation revealed during brain imaging research has raised doubts about a factual lack of consciousness or an inability to externally manifest own awareness (see Owen & Coleman, 2008).
552
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
time and rate) with a higher-order thought. What determines this synchronization? As it seems licit to disregard the possibility of being conscious of everything, and in lack of Kriegel’s clear position on the issue, it could be concluded that such synchronization is suggested to occur randomly and spontaneously. Despite any criticism, Kriegel’s COI hypothesis carries along some valuable contributions. In fact, while it seemed that he had disregarded Block’s taxonomy by treating consciousness as a unitary phenomenon, his NCC pattern through dlPFC could nevertheless fit with the occurrence of access- and self-consciousness. These aspects of our conscious experience could in fact justify the need of a structure able to support logical and critical thinking, planning, evaluative skills, and problem-solving abilities. Furthermore, Kriegel’s conceptualization of consciousness as the result of the mental act of reflecting on own, lower-level mental representations, runs closely to a possible definition of monitoring consciousness, and it further supports Block’s taxonomy. Indeed, Kriegel might have offered a valid hypothesis about the neural substrates correlating with the hierarchical construction of thoughts. Remarkable integrations to Kriegel’s theory would then be the attempt to explain the causal factors behind the emergence of consciousness (i.e. criteria behind the suggested neural synchronization), and the reasons for consciousness as a whole. 2.1.3. Crick and Koch: Tracing a framework for consciousness Based on the assumption that all kinds of conscious experience (visual, tactile, etc.) share a similar mechanism, and that therefore any knowledge gained about one perceptual modality could have been extended also to the processing of stimuli of different nature, Crick and Koch (1998) concentrated their efforts on the mechanisms that allow visual inputs to achieve conscious state. Crick and Koch’s theory followed the top-down approach suggested by Kriegel, in the sense that – in order to shape a proposal about consciousness’ locus/i and mechanisms – the authors took the step from their assumptions about what consciousness is, and why it manifests itself the way it does. In Crick and Koch’s understanding, consciousness is needed to select a single interpretation of a stimulus, rather than allowing the risk of different perceptual constructions leading to different – and possibly contrasting – behavioral responses. Compared with other animals, whose less complex environment grants behaviors to be guided by unconscious, reflex-like processing systems, humans can rely on two action modes, supposedly working in parallel: The on-line system (unconscious and rapid), and the seeing system (conscious and slow) (Crick & Koch, 1998, p. 98). Many action responses to sensory inputs are governed by the stereotyped scripts of the on-line system, whose adaptive value consists indeed in its resource-saving processing style. Concerning the nature of consciousness, Crick and Koch’s starting point was that consciousness is a private phenomenon, in the sense that conscious experiences are accessible only to the brain that produces them (Crick & Koch, 1998). Their belief in the inscrutability of conscious states guided then their hypothesis about the mechanisms behind the emergence of consciousness. In fact, Crick and Koch argued that stimuli are processed at stages of hierarchical sophistication (form, color, location, movement, and so on) and, at each stage, they are recoded and sent to the next level. The purpose of such recoding is that of translating for example the ‘‘color-input” into the ‘‘motor-output”, as we would do in pointing to a red light if we were asked to recognize it among others of different colors. The sequences of recoding from one representation system to another are beyond verbal descriptions (therefore, private), as they constitute the unconscious explicit representations9 exchanged between the neurons of a specific brain. The challenge at this point is to determine what are the conditions that allow a given stimulus to become conscious. Crick and Koch’s framework sets its center on the concept of competing coalitions, which refers to groups of neurons in competition one with another (Crick & Koch, 2003). Coalitions can vary both in size and in quality: A coalition produced by visual imagination might be less spread and less clear than one produced by a sustained visual input. Due to special circumstances, at any single moment one coalition wins above all others, and its contents are elevated to conscious states (Crick & Koch, 2003). The nature of such ‘special circumstances’ is however still quite speculative. Among the candidates, Crick and Koch name a particular kind of neural activity reaching a given threshold, or a specific type of firing among the neurons of the same coalition. Even more speculations will though meet the question concerning the factors that could determine either the particular neural activity or the occurrence of some specific type of firing. Crick and Koch propose in fact the occurrence of some special dynamic within the neurons, or perhaps the accumulation of chemicals such as Ca2+ in the neurons (Crick & Koch, 2003). During the competition phase, coalitions can then eventually be strengthen by the occurrence of synchronized firing, even though this event per se is not believed by Crick and Koch to be a sufficient condition for the shaping of a NCC.10 Another factor that could affect the competition among the possible coalitions is attention. In a more recent paper, Koch and Tsuchiya (2006) have further elaborated on the role of attention, which they propose as a mechanism independent from consciousness, and only able to influence its content, rather than to determine it.11 In fact, supported by dense experimental evidence, the two researchers brought up the argument that attention and consciousness, by clearly serving different purposes, must also rely on different neural pathways (see Koch & Tsuchiya, 2006 for an in-depth discussion about the differences between consciousness and attention). Following Crick and Koch’s framework, the representation reflected by the winning coalition will then likely reach the prefrontal areas where it will be made the object of conscious evaluation. It has to be noticed that
9 Explicit representations, in Crick and Koch’s terminology, are the products of small sets of neurons responsible for detecting specific features of a stimulus, such as color or shape. 10 Crick and Koch define the NCC as ‘‘the minimal set of neuronal events that gives rise to a specific aspect of conscious percept” (2003, p. 119). 11 Authors such as Dehaene and Nacchade (2001) assume that attention identifies consciousness since they argue that stimuli reach conscious state only if they become the objects of our attention.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
553
the representation reaching the prefrontal cortex reflects the computations of the explicit representations produced by the small sets of neurons. In such sense, Crick and Koch propose that consciousness is the result of a distributed system, involving some subcortical structures, and a variety of cortical areas12 whose explicit representations are joined together and become conscious states under the supervision of the prefrontal cortex. The above paragraphs have only attempted to present the broad margins of Crick and Koch’s framework for consciousness. The work carried on by the two researchers across decades is vast, both in terms of theoretical and empirical efforts. With no intentions, nor competence to offer an appraisal of the full value of such efforts, considerations shall in the present context limit themselves to the broad implications of their proposal. First of all, there is the need to appreciate the attempt to explain consciousness in terms of its evolutionary value. By understanding its purpose, we might find ourselves a step closer to understand how and where such purpose is fulfilled. Crick and Koch’s linking of consciousness to the organism’s processual needs also has additional relevance in terms of animal research. It could in fact be hypothesized that, proportionally with the complexity of its environment and higher the cognitive resources of the organism, the possibility to identify conscious behavior in non-human animals might turn into a successful enterprise. It seems in fact like there should be no reasons to assume that the proposed dynamic between competing coalitions should concern solely the way humans come to select the appropriate stimulus to address. Furthermore, Crick and Koch’s work lends itself to support at least three of Block’s categories of consciousness. The online system could in fact be responsible for modulating monitoring consciousness, while the seeing system (as concerning visual experience) could play a critical part in access-consciousness, by allowing a careful and detailed evaluation of the mental acts involved in the processing and response to a stimulus. The role of the prefrontal cortex in evaluating the inputs received from other brain structures could finally well accommodate the emergence of self-consciousness. On the other hand, Crick and Koch’s account appears to shed no light on phenomenal consciousness, nor on the reasons why there seems to be an emotional component to our conscious experience. If consciousness is about taking the right choice in a multifaceted environment, there might be an additional value in the emotional and qualitative charge that such evaluations carry along.13 A possible reason why Crick and Koch had – at least seemingly – ignored this element of our conscious experience could reside in an upfront bias in their research. In fact, they considered consciousness a composite phenomenon, where similar, but separated systems would determine each their specific results. In such sense, they assumed the presence of separate mechanisms behind the emergence of visual consciousness, self-consciousness, introspection, phenomenality, and so on (Crick & Koch, 1998, p. 97). By adopting such theoretical position, they seemingly allowed themselves to assume that the validity of a framework for consciousness would not have been affected just because it could not account for every aspect of the phenomenon. On the other hand, the fact of having left out of the theoretical whole an element as central as phenomenological consciousness, reveals perhaps one of the more noticeable gaps in Crick and Koch’s proposal. 2.1.4. Thagard and Aubie: Theory of emotional consciousness (EMOCON) Thagard and Aubie’s ambition has been to draw a theory which, taking the move from the assumption that emotions emerge from the interactions between various brain areas, can identify the neural mechanisms that cause emotional consciousness.14 According to the two researchers, there are five features intrinsic in all conscious experiences: Differentiation, integration, intensity, valence and change (Thagard & Aubie, 2008). Differentiation refers to the way our brain can produce a variety of different experiences e.g. sadness, happiness, depression, euphoria, etc.), which we are however able to distinguish one from another. Integration addresses instead our ability to enrich given perceptions, judgments and memories with a specific emotional quality (pleasantness when enjoying a favorite food, sadness when own football team looses, etc.). Intensity refers then to the degree of arousal that a given emotion can acquire. In such sense, there appears to be a quantitative difference between our feeling of happiness and euphoria, or between sadness and depression. Differences between various emotions also contain an element of valence, since we are in fact able to experience emotions quite at the extremes from each other (e.g. love and hate). Finally, change refers to the fact that the target of our attention can also determine the quality of our emotions. Whether my thoughts are directed toward an overdue bill, or about a soon-coming-up vacation can determine a shift between preoccupation to cheerfulness within seconds. According to Thagard and Aubie, a valid theory of consciousness should be able to explain how each of the above features emerges from our mental mechanisms. EMOCON consists of four basic elements – somatic perception, representation, cognitive appraisal, and working memory (WM) – which are at the core of four processual components. Concretely speaking, Thagard and Aubie propose that (1) somatic perception of a given object15 triggers the interaction between at least eight different brain areas (anterior cingulate, dorsolateral prefrontal cortex, orbitofrontal cortex, ventromedial prefrontal cortex, insula, amygdala, thalamus, dopamine system), (2) an appropriate representation of the object will consequently be generated by the specific firing of a determined population of neurons, (3) the representation will then receive a cognitive appraisal which (4) will depend upon the representation’s persistence in WM. Relying on these four components, Thagard and Aubie highlighted the value of their theory
12 They name the cerebral cortex, plus regions associated with it such as thalamus, claustrum, basal ganglia, cerebellum and a substantial number of brainstem projections. 13 This issue will be again considered in the next section of this article. 14 The term ‘‘emotional” is seemingly used to distinguish the feeling of experience from its otherwise basic cognitive processing. 15 Thagard and Aubie assume that this mechanism concerns also to the way specific populations of neurons react to internal events. It is compelling to note here the contribution of Damasio’s inner sense proposal. Damasio’s theory is discussed later in this section.
554
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
by showing that these could account for the aforementioned five features of emotional experience. In such sense, integration (explained by component n.1) appears to be the implicit result of the interaction between the many brain areas supposedly involved in emotional consciousness, which respectively cover various roles, such as stimuli detection, meaning attribution, memory. Emotional intensity (explained by component n.2) will instead be accounted for in terms of firing rates in the neural population correlating with the given object. The effects of amphetamines are brought up by the authors as evidence to show how increased firing in dopaminergic systems determines an enhancement of emotional intensity (Thagard & Aubie, 2008). The 2nd component of Thagard and Aubie’s theory is also called into explain valence in terms of changes in dopamine system and in the level of activation in the prefrontal cortex. These shifts will determine the occurrence of positive or negative emotions, depending on an increased or decreased activation. On such issue, the authors refer to experiments where manipulation of brain activation through deep brain stimulation or drugs determines emotional changes in the subject (Thagard & Aubie, 2008). Differentiation (explained by component n.3) is then proposed to be sustained by the cognitive appraisal of the representation. In other words, our experience of happiness rather than sadness relies on our cognitive evaluation of our specific emotional state. The 5th feature of emotions referred finally to the ability to change the quality of our emotions. This mechanism (explained by component n.4) is suggested to be relying on WM. Through its underlining features (recurrent activation, decay, stimulation and inhibition), WM can in fact determine the sustenance or restraint of a given population’s activity, modulating the emotional shift between different stimuli. In sum, according to Thagard and Aubie’s theory of emotional consciousness, an emotion is not the cognitive evaluation of an event, nor the perception of a stimulus: An emotion is a pattern of neural activity which involves all the brain areas indicated in the EMOCON model. On the other hand, emotional consciousness is not the product of any of the computations occurring within and between any structure in the brain, but ‘‘emotional consciousness just is the overall neural process that takes place in the interacting brain areas” (Thagard & Aubie, 2008, p. 818). It appears that EMOCON possesses many of the characteristics auspicious for a theory of consciousness. Its main quality is perhaps that of having proposed consciousness as involving a broad array of brain structures, rather than being localized in a specific structure (see Kriegel’s theory, earlier in this section). Its being a distributed system can in fact support the qualitative difference between the four forms of consciousness proposed by Block (1991). Thalamus and prefrontal structures could in fact possibly be involved in access and monitoring consciousness,16 while amygdala and insula could likely be modulating self- and phenomenal consciousness.17 Unfortunately though, it seems like Thagard and Aubie’s proposal lacks an important explanation: In fact, unless we are to assume that everything that impacts our internal and external sensors is to produce a detectable emotion, we have not been told on which basis the emotional selection occurs. In other words, how come we can be aware of the emotions generated by some stimuli, while others seem to produce no noticeable effects? Thagard and Aubie would perhaps not consider such question since, by regarding emotional consciousness as the ‘‘overall neural process” between perceptual and cognitive mechanisms, they seem to imply that emotions are implicit in every stimulus processing. Beside sensing in this implication the risk for an emotional ‘roller-coaster’ of scarce adaptive value (shifting from sadness to happiness, then to preoccupation and so on, depending on whatever input fleetingly reaches our sense organs), it also appears that Thagard and Aubie might have left unexplained the instance of non-emotional cognitive appraisal. It might in fact seem possible to absorb the sight of a given object, say a cup on the desk, and be left emotionally blank by the percept, even though its features have been processed and its use appreciated. On the other hand, feelings can at times emerge in quite non-cognitive manners (e.g. ‘‘gut-feeling”18), or they can also appear as a consequence of other emotions (e.g. feeling generous as a consequence of feeling happy). It is difficult to guess why such experiences, although supposedly emerging from so broad neural pathways including most of prefrontal cortex, seem to catch us in a void of cognitive awareness. A possibility could be to assume that, while cognitive processes can also occur unconsciously, the emotions that would accompany such neural patterns would still be detectable. Unfortunately, Thagard and Aubie have in their theory – to our knowledge – not devoted any concern to unconscious mechanisms, regardless of these withholding feelings or cognition. More importantly, they have not offered any suggestion about what the overall purpose of emotional consciousness should be, and therefore allowing the doubt about whether it is a mechanism implicit in all neural processes of all animals, or something that has evolved for some precise reasons in human-kind. As a concluding note in favor of Thagard and Aubie, it should be noted that their theory has offered a further challenge to Dualism, by showing how stimulus processing and consciousness’ emergence could in fact be two aspects of a same mental act. This point was much stressed by the two authors who stated that, ‘‘brain activity causes emotional experience in addition to correlating with it” (Thagard & Aubie, 2008, p. 819). 2.1.5. Damasio: The biological nature of consciousness While Thagard and Aubie have shaped their account on the assumption that emotions are the emergent property of any cognitive elaboration, Damasio gives emotions a causal role in the emergence of conscious experience. Damasio’s theory of consciousness is based on three fundamental elements: Emotion, feeling and feeling of feelings (Damasio, 1999).
16 The thalamus involvement could be determined by its function as integrative node between different brain areas processing external perceptions (Thagard & Aubie, 2008, p. 820), while prefrontal lobes’ role in working memory and decision making would appear to make such areas potential co-modulators of both access and monitoring consciousness. 17 Both the emotional component modulated by the amygdala, and the processing of somatic perception investing the insula could indeed accommodate the profile of phenomenal and self-consciousness. 18 The ‘gut-feeling’-phenomenon will be further discussed in the next section.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
555
Emotion, in Damasio’s application of the term, indicates the unconscious neural reaction triggered by a given stimulus and involving a broad range of brain sites. Specifically, he proposes that the ‘emotion-inducer’ sites are mostly situated in the subcortical nuclei of the brain stem, hypothalamus, basal forebrain and amygdala (Damasio, 1999, pp. 60–61). Based on PET recordings, Damasio also informs to have identified different patterns of such activation with reference to different kinds of emotion (e.g. joy, sadness, anger, etc.). He found for example that sadness is consistently associated with activation in the ventromedial prefrontal cortex, hypothalamus and brain stem (Damasio, 1999). Feeling, the second element of his theory, is described by Damasio as the unconscious sensing of the neural changes triggered in the proto-self19 by the emotion. Supposedly, there are two kinds of neural substrates that generate feelings: The first refers to the changes related to bodily status, which are mediated by either chemical messages (via bloodstream) or by electrochemical signals (via nerve paths); the second neural substrate mediates cognitive changes, which depends on the release of chemical substances in the nuclei of basal forebrain, hypothalamus and brain stem. Specific alterations due to these localized activations will finally determine a particular type of behavior, such as bonding, playing, or inhibition of body signals (Damasio, 1999). Damasio argues that the neural alterations caused by the emergence of feelings are also responsible for changes in the quality of the organism’s perception (e.g. visual detection ability, acoustic acuity, etc.). The final element of Damasio’s theory is represented by the feeling of feeling, also called core consciousness, which is the result of the correlation between the individual’s perception of the changes occurred in the proto-self and the neural activation produced by the emotion-inducer site. The neural substrates for core consciousness are supposedly situated in the cingulate cortex which, to merit of its many subregions and extended connections with the somato-sensory cortex, can receive from thalamic projections and send forward to higher-order structures in the inferotemporal and parietal regions (Damasio, 1999, in Bosse, Jonker, & Treur, 2008). Deploying its role as mediator, the cingulate could so be able to create a causal link between the appearance of the stimulus and the changes occurred to the body. The process from stimulus presentation to acquisition of consciousness crosses, according to Damasio, five stages: (1) A stimulus (an inducer of emotion) impacts the sense organ. (2) The detection of the stimulus triggers specific neural nodes (emotion-inducer sites), which are specialized in the different classes of input, and located in the relative processing areas. (3) The activation of the emotion-inducer sites triggers consequently other signals toward other brain areas and toward the body. The responses that in turn will be started by the body and by the brain will constitute an emotion. (4) The proto-self will at this point change and the feeling will emerge. Note: Changes up to this point are called first-order neural maps and all occur unconsciously. (5) The changes in the proto-self and the activity in the emotion-inducer sites can eventually reach together the somatosensory area. If such instance occurs, they will then cross to the cingulate and be forwarded to higher-order structures in the inferotemporal and parietal regions. Consciousness will then emerge. Fig. 2 illustrates the dynamic of these mechanisms. As it is not within the aim of this article (nor competence of its author) to evaluate Damasio’s proposals about the anatomical loci of consciousness, considerations will be limited to some of the broad suggestions implicit in his framework. A first comment should concern the emergence of emotions. On the basis of brain imaging evidence, Damasio states that (1) given types of emotions (e.g. joy, anger, etc.) are determined by the activation of given neural patterns and (2) it is possible to identify the location of the various emotion-inducer sites. However, it could be questioned whether it is licit to assume that such activation implies causality, when in fact it could be a more modest sign of correlation. A step forward might eventually be the production of evidence showing that a specific neural pattern always and exclusively precedes the subject’s report of the specific emotion. Although still not drawing a straight line between the pattern and the emotion (other elements could still work in between), we could then at least discuss the presence of a causal connection. A further doubt will then invest Damasio’s power-concessions granted to automatisms in the unfolding of behavioral patterns such as bonding or playing. Since up to this point Damasio stresses that the whole constellation of events is occurring unconsciously (Damasio, 1999, p. 80), as determined by specific alterations caused by localized brain’s activations, it would appear that, whether or not we bond with our colleagues, is really far from issues of personality or social circumstances. As Damasio personally put it, ‘‘consciousness begins only thereafter that an organism that is responding beautifully to its environment begins to discover that it is responding beautifully to its environment” (Damasio, 1999, p. 283, own emphasis). Although it will here be agreed that – by being innate behaviors – bonding and playing might indeed be triggered by unconscious mechanisms, it will also be argued that they are likely modulated (at least in humans) by conscious behavior, and therefore do not occur unconsciously. In fact, much work (e.g. Reason, 1979) has been carried onto establish that it is for the most only familiar and relatively rehearsed action patterns that can be performed in automatic fashion. Admittedly, it could be looked with skepticism to the adaptive value of leasing into the hands of unconsciousness the duty to govern human social interactions. Finally, a breach could be sensed in Damasio’s theoretical tissue. He underlined that having a feeling is not the same as knowing a feeling, in the sense that, whether or not the processing of the stimulus will reach consciousness, the neural
19 The proto-self is the preconscious biological precedent of the sense of self: It is responsible for coding the physical structure of the organism. It is also called first-order collection of neural patterns (Damasio, 1999).
556
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Fig. 2. The model shows the chain of activation that follows the detection of the stimulus: Neural activation in the sense organ triggers signals from the emotion-inducer site toward other brain areas and toward the body, giving raise to an emotion. The emotion will then determine changes in the proto-self: The detection of these changes is what Damasio calls feeling. If the activation in the emotion-inducer site and the changes in the proto-self simultaneously reach the somato-sensory area, they will then be jointly forwarded toward the cingulate and consequently toward higher-order structures. Only at this point we will become conscious of the initial stimulus. If the matching of the signals from emotion-inducer sites and proto-self wont occur, the stimulus processing would take place unconsciously. The red arrows trace the route toward consciousness. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
changes leading to feeling will still take place. In other words, stimuli affect our brain activations, which in turn can trigger a worryingly broad range of behaviors (including social responses and interactions) without us being conscious of it. Again, the responsibilities placed by Damasio on our unconscious processes seem counter-intuitive and of blunt adaptive value. More so, he argues that consciousness is the result of changes which, occurred in both the emotion-inducer sites and in the protoself, simultaneously reach the somato-sensory area. It would have been relevant at this point for Damasio’s theory to suggest some of the possible criteria by which given stimuli become conscious, while others do not. Unfortunately, Damasio’s silence on this issue gives away an additional lack in his proposal: In fact, having left unanswered the question about what causes a stimulus to become conscious, he has consequently ignored the reason why it should become so. On a brighter side, Damasio’s detachment of the role of attention from the prerequisites for conscious experience appears to be a strong point in his account. Crick and Koch had also defended such dissection, presenting a rich empirical substrate to sustain their argument (see Crick and Koch’s account, earlier in this section). Furthermore, as Damasio implies that the mere exposure to stimulus is sufficient to trigger a chain reaction of mental activation, we are released from having to postulate some sort of attention device which, unconsciously and precognitively, should have determined which stimulus deserved our attendance.20 2.1.6. Baars: The Global Workspace Theory Baars’ Global Workspace Theory (GWT) (1988, 2002) is an attempt to explain the role of consciousness in the brain’s cognitive processes. Baars’ basic suggestion is that consciousness covers an integrative function: By creating access to the widespread range of brain resources, it enables communication between the various processual units in the brain. Without the intervention of consciousness, many brain functions (per se unconscious) would lack coordination, and critical elements of the organism’s mental efficiency would be disabled, such as learning, control over motor functions, unfolding of automatic action sequences, selective attention and the ability to introspect. In Baars’ own words, ‘‘consciousness is the gateway to the unconscious sources in the brain” (Baars and Gage (2007, p. 240)). Baars’ theory rests on three main constructs: The unconscious contexts, the conscious global workspace, and the unconscious specialized processors. Fig. 3 tries to illustrate Baars’ Global Workspace model. The unconscious contexts are coalitions of processors that together determine the kinds of information entering the workspace (based on the organism’s needs, motivation, etc.), and influence the interpretation assigned to these data (Baars &
20 A number of theories (e.g. Dehaene & Nacchade, 2001, Baars, 1988) suggest instead that input selection is mediated by attention, and that only at a subsequent point the stimulus is made the object of cognitive evaluation.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
557
Fig. 3. The model shows how input, after having seeped through the unconscious contexts, compete with each other to access the workspace. Inputs supported by multiple stimuli or producing greater degree of activation, may then receive particular conscious analysis due to an attentional spotlight. Inputs will then be forwarded toward the unconscious specialized processors from where they will either trigger an automatized response, or require the intervention of further processing and additional resources. In this later instance, a new input might be sent to the workspace from the specialized processors (red double-arrows in the above model). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Gage, 2007). Baars’ contexts could be described as the pre-existing archives of the individual’s personal expectations, assumptions, goals, and social and moral values. Unconscious contexts can though also be less stable dispositions of the person, such as an occasional mood or a personality state (Baars & Franklin, 2007). The conscious workspace (WS) is then presented by Baars as having the role to integrate and broadcast, eventually by means of a ‘‘spot-light” (selective attention), the information that has filtered through the unconscious contexts. Baars proposes that stimuli of different nature and content compete one with another to gain access to the workspace, and consequently to emerge into consciousness. In such sense, it is suggested that more inputs carry a same message, higher the chances for that message to be let in the workspace, and to be granted our conscious attendance. However, Baars also introduces the idea of attractors, which can be conceptualized as mnemonic nodes able to ‘‘pull” in the WS stimuli of specific contextual relevance (Baars & Franklin, 2007, p. 956). By his own admission, Baars has not been able yet to identify the brain correlates of the workspace: Nevertheless, he suggests that the reticular and intralaminar nuclei of the thalamus, and the long range cortico–cortico connections would likely be the best candidates (2007, p. 196). It is then interesting to notice that the characteristics of the workspace, especially considering the critical role by which it is invested, are at first sight disappointing: Limited capacity, consistency-rule (only inputs carrying not reciprocally conflicting contents can enter the workspace at a same moment in time), and low computational efficiency. Baars considers such limitations the price to pay for allowing that single, precise messages could be heard by a whole system (Baars & Franklin, 2007): Less flexibility to the gains of control and coordination. Finally, the term unconscious specialized processors can refer to a single cell or to entire networks of neurons: Their role is to process stimuli of distinct nature (visual, auditory, etc.), and to enable specific functions such as language decoding (Baars & Franklin, 2007). They achieve such goals by responding to the information broadcast in the workspace, and by performing a full-fledged cognitive analysis of such data. The result of such computations can then result in either one of two courses of events: (1) the processor ‘‘recognizes” the familiarity of the input and triggers an automatic response-behavior (without consciousness’ involvement), or (2) the stimulus demands further analysis, and there is therefore the need to recruit more specialized processors. In the latter instance, a new message will be broadcast into the workspace posting the request for additional processual help: Marked by the emergence of the attentional ‘‘spot-light”, a coalition could then be formed to attend to the specific task. In contrast with the characteristics of the workspace, the specialized processors are supposedly able to rely on a much larger processual capacity, no consistency demands, and great computational efficiency (Baars & Franklin, 2007, p. 958).
558
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Baars’ theory of consciousness has received support by a non-negligible number of researchers, who have further elaborated his workspace model (e.g. Dehaene & Nacchade, 2001), and added to its support interesting evidence (see Baars and Franklin (2007) for a cognitive architecture of the model; Wegner, 2002). It is in fact undeniable that Baars’ theory presents a number of well formulated hypotheses, which could indeed account for some features of consciousness. First of all, it should be mentioned that the conceptualization of a global broadcasting appears indeed very plausible, given the topographical arrangement of brain structures and of their functions (see also Baars & Gage, 2007). On the other hand though, some scientific (and human) uneasiness might emerge in conceptualizing consciousness as a phenomenon guided and constrained by unconscious mechanisms, and in thinking about the workspace as a passive spectator of mental processes. Besides the feeling of lacking control over a great part of what goes on in our mind, it could be wondered about the risks of leaving matters as important as the selection of environmental stimuli in the hands of unconscious activation thresholds. Furthermore, it is not made clear how such selection can indeed take place. In fact, since Baars’ model reposes stimuli processing only after their existence has been broadcast in the conscious workspace, selection of stimuli appears to occur in absence of cognition. On a different note, Baars’ intuition to link the emergence of consciousness to working memory appears of most remarkable value. In fact, by allowing the selection of a stimulus not to be triggered exclusively on the basis of its own features, but also by top-down processes, Baars has possibly explained how consciousness about the absence of a stimulus can sometimes unexpectedly reach us. An example could be the way in which, upon entering a crowded restaurant, we could become suddenly aware of an unexpected silence. The consciousness of such stimulus-absence (that is, people’s voices, which are part of the mnemonic ‘restaurant-schema’) would indeed likely hit us even if our attention had up to that point been focused on other thoughts. On the other hand, Baars still has not completely unchained the triggering of any conscious state from the stimulus physical reality, let this being absence or presence. It could therefore be an improving factor in a theoretical account of consciousness to hypothesize a mental mechanism which operations wont depend on external stimuli: As a start, it could account for the presence of some elements of consciousness identified in patients in vegetative state (Owen & Coleman, 2008). This topic will be discussed again in the next section. A final note should be directed to the fact that Baars’ theory has more certainly lent theoretical space to Block’s taxonomy of consciousness. Access-consciousness and self-consciousness would in fact seem the direct results of the information broadcasting: Data, while possibly not having been centred by the attentional spotlight, are however available within the conscious arena for us to reflect upon them. Monitoring consciousness can also find its explanation in the way by which familiar information are automatically forwarded out of the workspace and through the receiving processors responsible for coordinating the output. Unfortunately though, Baars’ theory of consciousness does not offer any enlightenment about phenomenal consciousness’ mechanisms, nor about its purposes. 2.2. General considerations There could be no argument against the limits of the review presented in this section: Many eminent accounts (e.g. Dehaene & Nacchade, 2001; Tononi & Edelman, 1998; Zeki & Bartels, 1999, just to mention a few) have been neglected, and much more could have been said about the theories actually presented. However, it was not this section’s ambition to trace an exhaustive landscape of the theoretical work that has built around the phenomenon of consciousness. The actual aim of the undertaken review was to offer a glimpse of the multiple, theoretical possibilities that can emerge by choosing different approaches to the original questions posed at the beginning of this section: What we mean by consciousness, how it emerges and what we need it for. The adoption of different initial conceptualizations has for example led authors like Baars to focus on the cognitive aspects and on the role of consciousness, while Thagard and Aubie’s interest for the emotional flavor of our experiences has determined their proposal of consciousness as a strictly phenomenological event, apparently not covering any functional role. Could we really assume that only one of these accounts has found the true heart of consciousness, while the other one is consequently mistaking? We should be very cautious about making such a choice. As Block’s work has outlined, consciousness is a multifaceted phenomenon and its presence, and therefore its role(s), are likely not to be pinned down on any single theoretical tissue. Put another way, this article will argue that any theory is likely to contain given elements of accuracy and, by highlighting the compatibility between the respective strengths, we might obtain conspicuous benefits. As a first, it should be more feasible for us to assess how far the knowledge we have jointly acquired can advance us in the hunt for consciousness. If each theory has shed a tiny spot of light on a single aspect of consciousness, all in all we might have indeed accumulated a respectable degree of illumination. Perhaps this was also the core of what Mysterianism21 meant when it referred to our necessity to dissect a phenomenon in smaller units of enquiry in order to understand it. By the same token, by joining our strengths, it should be easier for us to spot our weaknesses, and to consequently identify those issues mostly in need of refinement. Table 2 offers a composite profile of consciousness as it emerges by linking the points of respective strength and mutual compatibility between the philosophical accounts sketched in Table 1, and the neuroscientific theories considered in the section. However, the presented approaches have also exposed points of weakness by leaving behind some explanatory gaps, namely:
21
See Table 1.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
(a) (b) (c) (d) (e)
559
To explain what determines the emergence of consciousness of a stimulus but not of others. To explain emergence and function of phenomenal consciousness. To explain the factor(s) that allow a given emotion to become conscious. To explain why we should experience emotions at all. To describe a system able to simultaneously present distinct representations of each stimulus appearing at one time, while still conferring the unitary perception of the overall experience.
In the next section, a new theoretical framework will be introduced which, while respecting (and ad hoc further elaborating) the integrated profile outlined in Table 2, will also try to compensate for the theoretical vacancies exposed above. 3. Section 2 3.1. Endogenous Feedback Network theory This second section might hopefully gratify Ned Block’s recommendation suggesting that an effective theory of consciousness should: (1) allow space to phenomenal consciousness and consider its occurrence also in absence of access-consciousness; (2) offer a satisfactory account for consciousness as a whole, and do not limit itself to the conceptualization of the ‘workspace’; (3) account for much of the existing bulk of evidence. 3.2. On the conceptualization of consciousness The theory advocated in this section will propose consciousness as the phenomenon emerging from a distinct network of neural paths which shadows all mental processes. This distinct neural structure will be referred to using the denomination ‘‘Endogenous Feedback Network” (EFN). It is in fact its author’s suggestion that the EFN represents the self-generated feedback that the brain forwards to itself, and that we perceive in form of internal stimuli. The EFN, which suggested role is therefore to globally broadcast the occurrence of mental activations across the brain, can also be conceptualized as the reverberation of the brain’s computations. Following such understanding, consciousness is not depending on wakefulness, since indeed there should be no doubts about the occurrence of wide-spread mental activity in sleeping subjects.22 Accordingly, vegetative states should not preclude the emergence of consciousness, as in fact brain imaging techniques have revealed given degrees of mental activation in otherwise unresponsive patients.23 Consciousness wont be identified with awareness either: A significant part of our brain’s activities can in fact occur in lack of awareness (e.g. dreaming, reading, automatisms), but we still wont deny them being the results of our brain’s computations. By this same token, it should be licit to assume the occurrence of a reverberation of such computations also in absence of awareness. As Block has argued, there is a qualitative difference between access-, monitoring-, self-, and phenomenal consciousness: The present theory will propose that such distinctions correspond to different EFN’s outputs shadowing specific processual stages. The following subsections will present a nuanced analysis of these intrinsic components of consciousness. 3.2.1. Phenomenal consciousness Let’s say that you sit at your desk, concentrating on writing a text on your computer, when a sound, say the meowing of a cat, reaches your ears. The activation of the sensory receptors in your cochlea, triggered by the changes in air pressure caused by the acoustic stimulus, determines a neural change which activates the EFN. It should here be stressed that the role of the EFN’s signal, at this stage, will be to broadcast the fact that your ears have detected a sound. The EFN signal is therefore not triggered by the sound, rather by the neural activation that such sound has produced on the sensory receptors. At this specific stage, in the minimal time interval between the activation of your sensory receptors and the moment the input is tackled by your processing areas, you would therefore know that your ears have caught a sound, but you would not be able to say anything about the nature, meaning, nor location of it. At this stage of EFN activation, it is now suggested that you would be experiencing phenomenal consciousness: The experience of hearing, deprived of all references to the perceptual details specific to that particular stimulus. It is therefore this author’s opinion that one of the questions dearest to many philosophers, along the lines with ‘‘What is it like for me to taste coffee?”, will likely never obtain an answer. The phenomenality of an experience would belong in fact to a processual stage occurring before the involvement of our cognitive mechanisms: Basically, we cannot put words, nor conceive a thought, about an experience which in fact still lies at its preconceptual stage. By the time a meaningful idea can be generated about our experience, we will be expressing it in qualitative terms (good, bad, hot, red, far, etc.), which are determined by our sensory thresholds, by the evaluations of our specialized brain areas, and by our history of encounters with the given stimulus. In other words, our experience would have become full-fledged cognitive, loosing in this way its phenomenal character. A study carried on by Martinez and colleagues (1999) offers valid support to the occurrence of neural activations which, although linked to the presentation of a stimulus, are independent from attention and not involved in the stimulus’ cognitive
22 23
See Kozmová and Wolman (2006), Occhionero (2004), Campbell (2000), and Delacour (1997). See Owen and Coleman (2008) and Coleman (2007).
560
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
processing. Using both event-related potentials (ERPs) and fMRI, the researchers investigated subjects’ cortical activations during a stimulus discrimination task, obtaining in this way data reflecting both attended and non-attended stimuli. Martinez et al. were so able to identify an initial activation pattern, caught by the ERPs recording and common to both types of stimuli, taking place in the retinotopically mapped striate cortex ca. 50–55 ms after stimulus onset. No attentional changes appeared to have modulated such activation. The first trace of attention-related changes concerned in fact only attended stimuli, and it was revealed through fMRI mapping at 70–75 ms post stimulus onset in the extra-striate areas. These data are already quite interesting, since they place themselves directly against the hypothesis24 that attention modulates the initial passage of visual input from LGN through V1. But perhaps even more intriguing was the finding that, although the activity in V2, V3 and V4 appeared to determine increased activation in the striate (with onset 104–136 ms after stimulus presentation), ‘‘no attention-related changes were observed in the amplitude of the short-latency C1 component that reportedly represents the initial afferent response evoked in V1 by visual stimuli” (Martinez et al. 1999, p. 366). In their attempt to interpret the nature of such distinct activations, Martinez and colleagues hypothesized that the late (104–136 ms), attention-related activity in the striate was the result of the feedback from extra-striate areas back into V1. This re-entrant feedback supposedly determined the neural amplification of the original visual input. A second suggestion had been that the late activation in V1 reflected a top-down, attention–binding signal which did not modulate the initial, stimulus-evoked response. Further elaborating on Martinez’s interpretations, it is here proposed that the initial, distinct activation in V1 (50–55 ms post stimulus onset) reflected the activation of the EFN at sensory level. Not being involved in the processing of the stimulus, such activation remained unaffected by the attentional amplification determined by feedback from extra-striate areas. It will be further suggested that, while the early EFN signal crossing V1 was effectively caught by the ERPs, its trace – from extra striate back to V1 – could have got ‘masked’ by the increased activity generated by the processing of the stimulus. Although the recordings seemingly reported one pattern of activity from extra striate to V1 at 104–136 ms post stimulus onset, the EFN argument suggests instead the presence of two neural paths, closely paralleling each other, but unequivocally fueled by different sources. Respectively, such patterns would identify: (1) the top-down attention–binding signal (as proposed above by Martinez et al.) and (2) the EFN’s path accordingly triggered by such progressive processing of the stimulus. In further agreement with Martinez’s second hypothesis about the attention–binding signals, it is here suggested that the EFN’s early activity in V1, signaling a visual stimulus in a given visual field, could have triggered an automatic allocation of attention toward the originating site. As Martinez’s highlighted, ‘‘such a delayed attention effect was not evident in the present ERP recordings, but it could have escaped detection if the striate cortex source were weak enough to be masked by the stronger sources that were concurrently active in extra striate cortex” (Martinez et al. 1999, p. 367). By virtue of this hypothesis, the presence of a second and distinct neural path (i.e. EFN) could have escaped detection. Increased activity in V1, V2, V3, and V4 – modulating the stimulus processing – could in fact rapidly have hidden the EFN path which followed all such computations and that indirectly led to attention allocation. 3.2.2. Monitoring consciousness Moving back to our example of the meowing cat, as the acoustic stimulus reaches your temporal lobes and successive brain’s areas, it will be gradually processed, and then matched against the data present in your memory. At this stage, you should recognize the sound to be a meowing, you should be able to describe it in terms of loudness, pitch, etc., and you should have an idea about its location and distance with respect to where you are sitting. Along with the processing of the stimulus, it is here suggested that one among three possible routes will be activated: (1) You recognize your cat meowing and you associate the sound to it being hungry – Absentmindedly, you get up from your chair, go in the kitchen, open the fridge, take out a can of cat food, and empty it in its bowl. (2) You recognize the sound as a meowing, which is highly unusual since you don’t own a cat – You give to the sound your full attendance. (3) You process the meowing of a cat and it gets associated to your memory of already having fed it – Without being aware of doing so, you ignore the sound. The computations correlating with any of the above routes would be sustaining the activity in the EFN which, in the instance the stimulus had triggered route-1, would have determined the emergence of monitoring consciousness. Although only minimal cognitive resources will in fact be allocated to the familiar behavioral sequence (i.e. feeding your cat), it is proposed that the EFN will be monitoring your performance by continuously broadcasting the neural effects correlating with the relevant action patterns. If, while still engrossed in your thoughts, you were about to empty a can of peeled tomatoes in your cat’s bowl, specific cues (such as the label on the can or the bright red of its content) will likely produce a wrong match within the selected action schema. The EFN output will therefore broadcast the detection of such wrong match, and its alert would determine the sudden deviation of the stimulus toward route-2, with consequent attention allocation. It should be stressed here the implication that you had become aware of the fact you were about to make a mistake prior to your awareness of being about to pour tomatoes in the cat’s bowl. This later knowledge would in fact imply the more resource- and time-consuming processing of ‘‘tomatoes” in order to assess their inadequacy as cat food. Volz and von Cramon (2006)
24
See Crick (1984).
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
561
support the occurrence of such pseudo25-cognitive ability to evaluate own actions by proposing that ‘‘(. . .) people continuously, without conscious attention, recognize patterns in the stream of sensations that impinge upon them. (. . .) The result is a preliminary perception of coherence which the subject cannot yet describe explicitly” (pp. 2077–2078). The argument about a specific EFN output being behind the occurrence of monitoring consciousness, and therefore responsible for the detection of slips of action, can be supported by an interesting study carried on by Escera and colleagues (2003). The Spanish team sought to investigate the factors determining the triggering of an attention switch in subjects involved in a visual task performance. The central question in their study concerned whether the attention switch would be drawn by bottom-up activation occurring prior to proper processing of the target input, or whether instead the subjects’ attention would be activated by the semantic analysis of the stimulus. Participants in the study were asked to concentrate on the identification of visually presented even and odd numbers, while trying to ignore acoustic distractors, which had been categorized as identifiable/novel (IN), non-identifiable-novel (NIN) or repeating-standard sounds. Through a combination of ERP recordings and behavioral analysis methods, Escera and colleagues showed that, not only IN sounds consistently captured the subjects’ attention (as showed by an increased RT), but they also determined a novelty-P3 amplitude larger than the one produced for NIN sounds. The remarkable point to make here is that a number of studies (see Escera, Alho, Winkler, & Näätänen, R., 1998; Yago, Corral, & Escera, 2001) have already showed that P3 amplitude, which is traditionally associated with attentional allocation, increases proportionally to the saliency of the stimulus and to the degree of distraction by it implied. Even more intriguingly, while the N1 amplitude that followed the presentation of standard tones had appeared enhanced for novel sounds, no difference in such enhancement distinguished IN and NIN. It has to be noticed that a nonnegligible bulk of evidence has linked in these past years N1-enhancement to the call for focal attention (see Näätänen & Picton, 1987; Näätänen & Winkler, 1999). On the basis of the results gained through their study, Escera’s team was therefore able to draw the following conclusions: (1) although both IN and NIN stimuli triggered an identical call for focal attention (N1-enhancement), attentional allocation (P3) was only granted to IN stimuli; (2) the analysis of the semantic content of a stimulus did not occur automatically (in which case both types of stimuli would have produced similar P3); therefore, (3) attention allocation is the consequence of a covert attendance mechanism. These conclusions were used to support that, ‘‘underlying cerebral processes operate automatically in a bottom-up manner [. . .]. [. . .] Semantic analysis of significant sounds occurs after a transitory switch of attention toward the eliciting stimuli” (Escera et al., 2003, p. 2408, 2411, own emphasis). Such transitory switch had previously been considered by Näätänen (1992), who suggested that it ‘‘must be triggered by a transientdetector mechanism reacting to sudden changes in stimulus energy or by a change-detector mechanism” (Näätänen, 1992, in Escera et al., 1999, p. 2408). At the light of the EFN theory, Näätänen’s ‘change detector mechanism’ identifies the neural output determining monitoring consciousness. 3.2.3. Access-consciousness Having left you almost pouring peeled tomatoes into your cat’s bowl, we should now get back to the mechanism that, after the detection of the cognitive mis-match, would have determined the redirection of the stimulus’ response toward route-2 (see above). As the attentional spotlight got turned on, a full array of perceptual details had likely become available, both contextual (such as the content of the can in your hands, and the fact you were just about to make an error), and retroactive (e.g. the occurrence of the original stimulus, the meaning you attributed to it, and the fact you had remembered of not having fed the cat). The passage from low resources processing to full-blown attentional allocation has, in the terms of the EFN theory, modulated the shift from monitoring consciousness to access-consciousness. Furthermore, it is proposed that the abilities to ‘mentally replay’ sensory stimuli, and to contextually follow the footsteps of your mental machinery are brought to you by means of the EFN pattern unceasingly following your brain activity. It could then be argued that the access to such information could itself be determined by our cognitive mechanisms (perhaps as an element intrinsic in attention), and therefore the eventual presence of an underlying kind of consciousness would be redundant, and likely improbable. Let’s then get back to the meow, and let’s consider the case of you not owning a cat. As your brain works on different questions and possibilities (e.g. how a cat got into your apartment, how you could make it leave, and so on), you will also be aware of the taking place of such mental activities, and you will be able to follow them closely, carefully evaluating options. But most importantly, without having to dedicate to it an actual thought (that is, in lack of intentional cognitive activity and in absence of attention), at no point you will be in doubt as to whether you are still working on your text, or concentrating about a cat meowing.26 In other words, it is here suggested that our ability to access the contents of our brain’s elaborations follows a separate neural route from the one that modulates our ability to know that our brain is involved in such computations. While the first route is governed by cognition and modulated by attention, the second one is shaped by the EFN activation produced by the neural changes associated to such cognitive activity. In the terms of the EFN theory, this latter route identifies access-consciousness. Gratefully, the argument appears to accommodate two of the qualities of consciousness about which most philosophers,27 and a number of neuroscientists,28 have agreed upon: Immediacy and Intentionality. We are in fact immediately conscious of 25 The suffix ‘pseudo’ is used to indicate that, although relying on the participation of cognition, the mental act is performed with minimal resources’ expenditure and in lack of direct intentionality. 26 Zahavi (2005) strongly supported the inequivocability of such kind of inner knowledge. 27 E.g. Brentano (1874 in Thompson & Zahavi, 2007). 28 E.g. Kriegel (2007) and Thagard and Aubie (2008).
562
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
those percepts to which we have granted attention, without needing to intentionally form a thought to gain such awareness. More so, the content of our consciousness is always associated (intentional, Lat. in tendere, to tend toward) to a specific target, whether or not we produce a specific thought to bind such relationship. While cognitive mechanisms are then occupied in modulating the immediacy and intentionality of stimuli processing, we might need to consider the occurrence of access-consciousness as a parallel mechanism able to render such cognitive computations themselves the immediate and intentional objects of our consciousness. An interesting study performed by Kentridge Heywood, and Weiskrantz (2007) can help to further advocate the distinction between attention (as the outpost of cognitive processing), and our ability to access our cognitive processes. Cerebral achromatopsia is a neurological condition in which, due to damage to the ventromedial occipital cortex, the patient is no longer able to distinguish colors, to perform tasks involving color discrimination, nor for that matter to remember the very concept of seeing in color. Previous research conducted by Heywood, Kentridge, and Cowey (1998) had already showed that these patients were however able to distinguish between two targets (such as a square and its background) if these were equiluminantly colored (i.e. respectively, red and green). The patients’ knowledge could however not be accessed, in the sense that they actually could not tell in what consisted the difference between the square and its background, even though they knew that they differed somehow. Kentridge and colleagues built on Heywood’s results and, by manipulating factors such as luminance and chrominance, showed that achromatopsic patients can, besides distinguishing color borders, also distinguish between boarders of different colors. It is relevant to stress here a great difference between achromatopsia and other disorders such as blindsight. While in fact patients suffering from the latter disorder can report of not seeing a given stimulus – even thought they are able to react appropriately to it above statistical chance – achromatopsic patients admit of perceiving a difference between the objects, but they just cannot explain the nature of such difference. Kentridge, in a reply to a target letter written by Block (2007), proposed his own results as evidence for the differentiation between phenomenal and access-consciousness, suggesting that, ‘‘if [the process of color vision] is prevented from running to its conclusion, we might be left with an incomplete signal that gives rise to sensations, but cannot be integrated with cognition. Stimuli can therefore potentially elicit phenomenal experience [. . .] yet remain cognitively isolated” (Kentridge et al., 2007, p. 509). While agreeing with Kentridge’s final conclusion, it will here be further argued that, being his patients able to appropriately act on the targets (distinguish the shapes and report the detection of a difference), their color vision systems might after all have produced a relevant outcome, rather than a mere ‘‘sensation”. In this occurrence, the nature of their impairment could rest in an unresponsive EFN pattern which, by not broadcasting the achievements of the cognitive mechanisms, is impeding the emergence of access-consciousness. Congruently with this proposal, Baars had argued that the wide availability of stimulus information has to be considered determinant for cognitive access”. The results of Kentridge’s study offer considerable support to the suggestion of a distinction between cognitive mechanisms, specifically attention, and access-consciousness. However, such position is in clear disagreement with Block (1995), who had argued that cognitive accessibility is strictly determined by attention,29 and that attentional deficits, or limited processing time, would impair access-consciousness. In fact, in the cases reported by Kentridge, patients were under no time constraints, and they actually dedicated their full attention toward the study of the target. Furthermore, they were aware that the task aimed at investigating their ability to distinguish colors and, speculatively speaking, they might have been likely exerting intentional resources toward such goal. Still, the emergence of access-consciousness was to them denied. In sum, however we might choose to interpret the phenomenon denoted by Kentridge, it will still seem plausible to point to the presence of two distinct neural paths, respectively grounded in cognition and in consciousness. In fact, whether we suggest that a damaged cognitive system determines an equally impaired EFN signal, or whether we support the possibility that a damaged EFN path cannot broadcast the otherwise correct cognitive output, the fact that a stimulus, while cheating attention, can nevertheless reach us, still remains unscathed. 3.2.4. Non-cognitive consciousness Moving once again back to the example of the meowing cat, it has previously been indicated the possibility that the sound could have activated route-3 (see earlier this section), determining your apparent neglect of the stimulus. Based on the assumption that you had heard30 the sound, its inability to raise your attention,31 and eventually to trigger an access-consciousness trace, shows that the stimulus had been processed at least to the extent of assessing its nature, meaning, and relevance. In fact, had the sound been threatening or very unusual, your attention would have likely polarized toward it (see Dalton & Lavie, 2004; Escera et al., 2003; Parmentier, 2008). More so, its failure to initialize an automatic response pattern, determining so the absence of monitoring consciousness, might also testify that the stimulus had in fact found a match in your memory – in this case the knowledge of previously having fed the cat – which had then inhibited the ‘‘meowing > hunger > feed-the-cat” – automatism (see Verbruggen and Logan (2008) for a discussion about the effects of WM on automatisms). It will now be further
29
We could eventually agree on a mediating effect of attention toward access-consciousness. Hearing is here taken as an involuntary process: Granted that the sound is above a given threshold (to be detected by human ears or simply to overcome background noises), and that the individual’s hearing system is not impaired, the sound will produce an effect on the sensory system. The question is therefore whether it will stimulate the individual’s attention or not (e.g. Cocktail party effect, see Moray, 1959). 31 Instances of attentional deafness, where a stimulus gets neglected due to specific circumstances (such as mental resources being intensely allocated on an ongoing task), are certainly possible. However, and we will get back on this issue later in this section, in this article it is argued that attentional allocation to such stimuli might be delayed, rather than aborted. 30
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
563
suggested that the occurrence of such wide-spread mental activity triggered by the sound, although it had not attracted your overt form of attendance, had nevertheless been detected and broadcast through the EFN, determining what we will call noncognitive consciousness. The reason for dismissing the term ‘unconsciousness’ lies in the linguistic implication that unconsciousness resides outside the boundaries of consciousness, and in such sense it seems to identify a separate element. Along the lines of the EFN theory, the distinction placed between other kinds of consciousness and non-cognitive consciousness rests instead on the fact that the latter shadows the processing of a kind of stimuli which, at the time of exposure, cannot (or should not) receive cognitive enlightenment. According to such interpretation, non-cognitive consciousness is therefore still the result of a specific type of EFN activation, and it should consequently join phenomenal, access-, monitoring- and self-consciousness in a complete taxonomy of the phenomenon. Based on its preliminary evaluation,32 a stimulus could in fact be categorized as ‘‘scarcely relevant/irrelevant”, ‘‘non-identifiable”, or perhaps ‘‘too unpleasant”, and consequently placed in a state of cognitive isolation. The trace of its passage might nevertheless affect your knowledge (see Kouider and Dehaene (2007) for an in-depth review of visual masking). Accordingly, it is here hypothesized that, even having not acknowledged the meowing, if you were asked whether your cat was in the apartment or out in the garden, your answer would have most likely been correct. Depending on specific circumstances, the non-cognitively processed stimulus could then eventually be elevated to full cognitive status, marking the passage from non-cognitive consciousness to access- or monitoring consciousness. The following ones are examples of such circumstances: (1) The mnemonic trace of the stimulus may link to newly acquired data, and its previous detection will then tag along the trail of the new one. Ex.: Learning about a robbery at the bank at the corner, and realizing of having seen that same day a blue car parked outside the building; (see Nelson and Goodmon (2004) for a discussion about implicitly activated memories and the role of context). (2) As the level of our available cognitive resources shifts, the stimulus that had previously not been granted attention might be reconsidered. Ex.: Having finally terminated to write your article, you might unexpectedly remember that the cat had been meowing; (see Meyer & Kieras, 1997; see also Navon and Miller (2002) for a discussion about queuing process). (3) As new circumstances might have affected its relevance, the stimulus previously unaddressed might suddenly gain overt attendance. Ex.: Driving and noticing the rain on the car’s wind shield, when suddenly the mental ‘sight’ of your bedroom’s window having been left open emerges in your mind. All above examples can be supported by Geiselman’s et al. (1985) Cognitive Interview33 technique which, through appropriate priming, can enable subjects to recall details of a scene which they actually were not aware of having caught. The suggestion in this context is that, although the trace of the detection and of the stimulus’ initial processing has been broadcast by the EFN and it is therefore present in non-cognitive consciousness, reportability is only possible through cognitive activation. Put another way, subjects who are non-cognitively conscious of having detected and processed given percepts, cannot intentionally gain access to their contents. Such percepts might however emerge by linking to those that have instead activated either a monitoring- or an access-consciousness trace. This paper’s proposal about the non-cognitive aspect of consciousness appears further compatible with our dreaming experiences during which, supposedly since our cognitive abilities appear qualitatively reduced (speech, reasoning, planning, etc.), the encoding and storing of our mental artifacts (dreams, sleep talking, reactions to external perceptions such as movement or temperature) appear to lack a given degree of effectiveness. It is our present suggestion that the mental activation generated during sleep is accompanied by a pattern in non-cognitive consciousness. Upon awakening, appropriate priming might in fact bound such trace to cognition allowing us to recall our nightly percepts. A study considering the incidence of dreams’ recall in patients suffering from alexithymia34 can accommodate the suggestion of a strictly non-cognitive trace following our oneiric activity. De Gennaro et al. (2003) have in fact showed that their patients presented significantly greater difficulties in recalling their dreams than a control group of healthy patients. It is here hypothesized that the emotional impairment caused by alexithymia might in fact further obstacle any cognitive retrieval by offering an initially impoverished non-cognitive trace. Studies based on classic fear conditioning paradigm offer further evidence to the EFN theoretical arguments. In their backward masking paradigm, Öhman and Soares (1993), Öhman and Soares (1994) paired fear-evoking and pleasant stimuli to neutral ones. FMRIs recording obtained while the subjects viewed the neutral stimuli which had previously been paired with the fear-evoking ones, showed increased brain activation and heightened physiological arousal, compared to the neural effect gained by neutral stimuli previously paired with pleasant images. We will now borrow such results to argue that, while the cognitive trace of the fear-evoking stimulus had been inhibited by the stimulus that had followed, its non-cognitive trace had nevertheless been broadcast and, by means of the cognitive priming (the neutral stimulus), such unreportable input was still allowed to affect the subjects’ arousal.
32 It is referred here to the kind of evaluation taking place prior the eventual granting of attention (see Escera et al., 2003; Näätänen, 1992, mentioned earlier this section). 33 The method is based on the evidence that memory retrievals are facilitated by cues: By encouraging the subjects to mentally reinstate the context of events, to report all possible details, to change perspective and reporting order, the chance of stimulating new cues will increase. 34 Alexithymia is a disorder characterized by a strong impairment to identify, distinguish and describe own and other’s feelings, by barren fantasy, and by a cognitive style strictly oriented toward the processing of external stimuli.
564
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
In sum, it is hereby argued that non-cognitive consciousness is characterized by particularly low activation levels, congruent with the minimal cognitive resource allocation granted to the processing of those stimuli of which it follows the trace. Cognitive access to these stimuli occurs then only if they become the target of specific priming, which could constitute a sort of stimulus’ neural amplification. Depending on their relevance and/or on other contextual factors,35 traces in non-cognitive consciousness will eventually decay. Ned Block, reviewing a number of studies pivoting on stimulus detection paradigms, concluded that ‘‘experiential content36 [. . .] can be instantiated without the kind of access that is based in the Access NCC” (Block, 1995, p. 49). Block’s definition of NCC reflected Crick and Koch’s, that is ‘‘[NCC is] the minimal set of neuronal events and mechanisms jointly sufficient for a specific conscious percept” (in Block (1995, p. 46)). Both above-mentioned quotes will therefore be taken as support for the admissibility of a specific neural path that, while distinguishing itself from other forms of consciousness – phenomenal, access, monitoring and self-consciousness – is singularly able to determine a percept. Following such argument, the NCC underlining route-3 (see previously this section) identifies non-cognitive consciousness. 3.2.5. Self-consciousness As it has been described in the above paragraphs, a stimulus – depending on its preliminary assessment – can activate either an automatic action sequence, or it can demand our attentional allocation, or eventually fail to produce any overt reaction. In each of these cases, the activation produced by the stimulus will respectively determine the emergence of monitoring-, access-, or non-cognitive consciousness. It is relevant to stress the fact that, based on the EFN’s assumptions, only the activation of an automatism or the allocation of attention will lead to a response which we will be able to cognitively trace, both in terms of its determining causes and of its possible consequences. In the (by now historical) example of the meowing cat, if the sound would have triggered the ‘‘feed-the-cat”-automatism, you would have performed the necessary steps, implicitly knowing what you were doing and, later that day, you could have likely been able to recall having got up and fed the complaining pet. Even more probable would have been your chance of remembering having heard the sound, if its peculiarity would have demanded your full attention (route-2). It will now be suggested that your ability to cognitively trace both types of response identifies what we will call self-awareness. In other words, your action responses will become the objects of a new processing cycle, giving so raise to your awareness of your physical movements (getting up from the chair, going to the kitchen, etc.), awareness of your movements’ goal (i.e. feed the cat), and awareness of your actions being in response to your cat’s previous meowing. The EFN theory further proposes that, while the outputs of our cognitive systems contribute to the emergence of selfawareness, the activation reported respectively by phenomenal, access-, monitoring-, and non-cognitive consciousness merge together in the flow of our self-consciousness.37 In other words, it is argued that self-consciousness identifies the EFN pattern responsible for broadcasting the overall effect of the activations reported by every NCC occurring at any one time in our brain. As the constantly circulating EFN’s core-patterns become the inputs for new mental acts (own perceptions bring processed as new stimuli), the individual-as-object, reported through self-awareness, is paralleled by the individual-as-subject through the emergence of self-consciousness. In sum, according to the EFN theory, while self-awareness identifies our cognitive knowledge about performing and evaluating specific actions, self-consciousness identifies our conscious knowledge about conceiving and experiencing such acts. These suggestions can then accommodate the duplicity intrinsic in our experiences. As we carry on an action, we can be aware of our performance, appraise its contextual appropriateness, and its general quality (self-awareness). Simultaneously, we are able to distinctly assign to ourselves the ownership of our acts, and to recognize as our own the experiential outcome that might emerge from them (self-consciousness). Furthermore, an additional feature distinguishes self-awareness from self-consciousness. In fact, by receiving the cumulative outputs of phenomenal, monitoring-, access-, and non-cognitive consciousness38 (which are the instantaneous products of on-going activations), self-consciousness is strictly bound into the present. While I may now still be aware of, say, having come late to the dinner-party I attended to yesterday, I can no longer directly39 experience the embarrassing self-consciousness of feeling the eyes of the other guests fixed on me. It is not an easy task to find existing evidence in literature to back up the suggestion that self-awareness and self-consciousness are terms intrinsically referring to two separate phenomena. In fact, it is extremely common to find that authors, when wishing to indicate disorders of self-perception, use the two locutions as synonymous of each other. This practice makes it hard, if not impossible, to distinguish the neural profile of the respective mechanisms in otherwise very valuable studies (e.g. Hobson, Chidambi, Lee, & Meyer 2006; Laureys, Perrin, & Brédart 2007; Lutz, 2007; Tsakiris, Hesse, Boy, Haggard, & Fink, 2007). However, if we were indeed to adopt the case of self awareness and self-consciousness being the results of two distinct neural mechanisms, we should then also assume the possibility of cases in which, due to damage or to other organic 35 The individual’s level of dispensable resources or the very nature of the stimulus (whether or not it might represent a threat to the individual’s psychological well-being) could represent examples of such influential variables. 36 It should be underlined that the presence of a ‘content’ shapes a clear differentiation between the specific profile of non-cognitive consciousness and that of phenomenal consciousness. In the terms of the present paper, the latter lacks in fact any factual content. 37 This hierarchical organization of consciousness borrows from Block’s suggestion concerning the presence of various core NCCs and total NCC (see previous section). 38 The contribution of non-cognitive consciousness into self-consciousness is intended to be strictly covert: It would therefore accommodate instances where, although not able to explain the causes, we become self-conscious of given mental states, such as anxiety, nervousness, confidence. 39 By efforts of memory and through context reinstatement (that is, through cognitive reappraisal), I might though be able to recreate the indirect experience of yesterday percepts and therefore to trigger the relative emotional state.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
565
abnormalities, only one of the two patterns could result impaired. Schizophrenia appears then to fit quite appropriately such instance. In a very interesting paper written by Sass and Parnas (2003), schizophrenia is described as a disorder of the self, especially in terms of hyper reflexivity and diminished self-affection.40 As reported by the authors, patients develop an exaggerated tendency toward self-analysis, transforming the self in an object of constant and detailed observation (hyper reflexivity). The perception of having become a spectator of oneself, combined with the diminished belief of being able to have an effect on oneself (self-affection), determines the gradual decline of the individual’s sense of differentiation between the self and others, until the patient comes to perceive him/herself alien and externally located (Sass & Parnas, 2003). As the condition progresses, patients begin to experience the so-called thought aloud and running commentaries. Both symptoms refer to auditory hallucinations, where the former are usually perceived as two or three different voices discussing and evaluating the patient’s actions, while the latter is often a single ‘‘entity” describing the patient’s ongoing acts (Sass & Parnas, 2003). Many researchers (e.g. Cahill & Frith, 1996; David, 1999) support the thesis that the voices brought about by schizophrenia actually reflect the patient’s own inner voices. In fact, the absence of causal connections, the lack of time and place definitions, and the fact that the speaker’s identity seems implicit, are reportedly all elements typical of inner language (Cahill & Frith, 1996). Additionally, the patient affected by schizophrenia reports a loss of spontaneous movements and of schema controlled automatic processes (Gray, 1991, in Sass and Parnas (2003)), together with an obsessive and unstoppable tendency to over-think every act, object and percept, with a consequent overload that impairs the mental machinery in its all (Frith, 1979). In Antonin Artaud’s41 words, ‘‘The brain sees the whole thought at once with all its circumstances, and it also sees all the points of view it could take and all the forms with which it could invest them, a vast juxtaposition of concepts” (in Sass & Parnas, 2003, p. 436). It has previously been suggested that self-consciousness mediates the global feedback comprising the signals produced by the specialized neural clusters modulating phenomenal, monitoring-, access, and non-cognitive consciousness. The purpose of self-consciousness would therefore be to present the brain with the comprehensive and integrated map of all ongoing activations occurring at any point in time, shaping in this way the causal background of the individual’s states and responses. At the light of the EFN’s assumptions, it is therefore proposed that the clinical profile of schizophrenia is compatible with the disruption of a normal flow of self-consciousness. In fact, the patient is aware of being the performer of given acts, and s/he is cognitively aware of own perceptions (e.g. voices, words’ meanings, etc.). However, the individual seems to lack any element to tie those percepts to the workings of own mental mechanisms. In symmetry with such proposal, Frith (1992) suggested that ‘‘schizophrenia can be explained as a neurophysiologically-based decline in the feedback” (in Sass & Parnas, 2003, p. 431), while Llinas, Ribary, Joliot, and Wang (1994) described the hallucinations and delusions characteristic of the condition as ‘‘derangements of the ability to distinguish exogenous from endogenous activity” (in Vaitl et al., 2005, p. 112). Compatibly, Kircher and Leube (2003) has reported studies which argued that conditions such as schizophrenia are caused by deficits in self-monitoring, with the consequent inability to distinguish the effects of self-generated actions and thoughts from those performed and initiated by others. Following the EFN proposal, self-consciousness is the phenomenon that modulates such ability to distinguish between the two categories of percepts. In fact, the perception of self-generated acts will emerge carrying along a trace of previous mental activations which confers to them subjective historical consistency and causality. In contraposition, the perception of other’s actions would appear clear from any personal and historical dimensions, as it will be the result of self-contained sensory activations. The need to assume the presence of an underlying system able to report back to the brain its responsibility in self-generated acts had emerged from the PET recordings registered by (Decety et al. (1997), in Georgieff & Jeannerod, 1998). The researchers showed in fact that, whether subjects were asked to perform a simple manual task or merely to observe other participants doing the same, there would be great overlapping between the brain areas becoming active in both conditions.42 The question that followed such results concerned then the way in which the brain could ‘‘feel” the difference between the two similar activation patterns. It seemed in fact clear to Decety and colleagues that to code for a motor response and to gain perceptual experience of it must not be part of a same act. Such was also the conclusion reached by Castiello, Paulignan, and Jeannerod (1991), who showed that the two processes can in fact be separated. In their study, they asked subjects to track by hand a target which would unexpectedly change direction. The trajectory modification of the subjects’ hands occurred 100 ms following the target’s leap, but the participants reported the perception of their hand’s movement at least 200 ms later. Considering such evidence, Georgieff and Jeannerod (1998) concluded: ‘‘Results suggest the existence of a double coding of action-related information. Signals used for controlling motor execution would be different from those used for generating conscious judgments on an action” (p. 470). Aiming at testing the degree of ownership that schizophrenic patients would report about voluntary movements performed with a joystick, Spence et al. (1997) revealed then that the delayed perception of own motor output reported by Castiello’s healthy subjects, had almost completely dissolved in delusional patients. In fact, most participants experienced a sort of alien control during the performance of the task (in Georgieff and Jeannerod, p. 473). The assumption that a dysfunction in self-consciousness is behind the schizophrenic chart is then most concretely supported by Georgieff and Jeannerod, who
40
The authors stress the fact that the term affection in this context does not indicate fondness, rather ‘‘the ability to affect” (Sass & Parnas, 2003, p. 428). French poet and essayist, Artaud (1896–1948) described in his writings his personal experience with schizophrenia. His self reports have offered great insights especially in the negative symptoms of the condition. 42 Rizzolatti, Fadiga, Gallese, and Fogassi (1996) had studied this phenomenon in monkeys where the activation (ascribed to ‘‘mirror neurons”) had been attributed to the ability to recognize others’ motor actions by comparing them to similar ones coded in own motor cortex. 41
566
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Fig. 4. When considering the emergence of consciousness, theories tend to polarize toward specific hypothetical stances: Consciousness can consequently be proposed as the result of specific computations or representation occurring either anywhere in the brain or in specifically dedicated brain areas. Alternatively, conscious states can emerge from non-specific computations or representations though occurring in dedicated brain structures.
stated that: ‘‘Self-consciousness does not rely on discriminating between central signals and sensory reafferences, but on discriminating between central representations activated from within and those activated by external agents” (1998, p. 474). 3.3. On the emergence of consciousness At the beginning of the previous section, it has been outlined Atkinson’s suggestions about the different hypothetical stances intrinsic in the attempts to address consciousness’ emergence. Fig. 4 sums up the possible combinations. This paper will advocate for great overlapping between the various theoretical stances, rather than more rigidly take them as separated epistemological assumptions behind the emergence of consciousness. In such sense, it will begin by suggesting that the specificity possessed by the computations giving raise to consciousness is determined by the fact that they occur within a specific kind of neural net which, rather than concerning itself with the processing of any stimulus, has its object in the processual machinery itself. Put another way, it is here argued that the neural effects of non-specific43 computations triggered by stimuli processing acquire specificity as they impact the EFN, consequently obtaining widespread neural broadcasting. It is then further proposed that the computations occurring along such specific neural paths can give raise to specific representations: Such specificity consists in the fact that we are allowed to perceive them as internally generated. In other words, mental representations become specific due to the broadcasting of the neural changes caused by the detection and processing of the stimuli that shape them. Finally, the EFN theory argues that consciousness is the result of computations which, while occurring within a distinct neural structure (EFN), can take place anywhere in the brain. This last feature is consequent to the proposal that consciousness identifies the effect of activations occurring at any one time everywhere in the brain. This paper does not hold the presumption to be presently able to offer a qualified and detailed account of the neural mechanisms leading to the emergence of consciousness which, beside any issue of personal competence of its author, is beyond the scope of this present work. Aiming in fact at offering a general framework for consciousness, it is likely more relevant in this context to plausibly support the suggestion of consciousness emerging from a distinct neural network, shadowing all mental processes. The proposal of consciousness arising from such distributed network appeared already implicit in Lawrie et al. (2002) conceptualization of disorders of consciousness, described as, ‘‘alterations of consciousness [that] arise from defective connectivity or from defective interactions between distributed brain regions” (in Vaitl et al., 2005, p. 111). Tononi and Edelman (2000), focusing on the neural elements behind such connectivity, highlighted the link between consciousness emergence and thalamo-cortical binding. The researchers argued that conscious experience emerges as the separate neuronal groups in the thalamo-cortical system (taken to identify different states) bind together following recurrent interactions, shaping in this manner what Vaitl called ‘‘an integrated neural process or functional cluster” (2005, p. 112). Llinás and Paré (1991) have pointedly offered evidence for gamma oscillations at 40 Hz as being the neural spark toward the binding underlying conscious
43 Note that the term spec.-/non-spec. computation is here meant to indicate the possession or lack of specific characteristics enabling a given computation to become part of the stream of consciousness. The intrinsic specificity of neural events based on their nature (e.g. visual, acoustic, etc.), location and mechanism is of course beyond any contextual discussion.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
567
Fig. 5. The drawing shows the mechanisms of recurrent interactions assumed to be behind the emergence of consciousness. The activation produced by a stimulus on the neurons in the cognitive path (blue) triggers activity in the neurons compounding the consciousness network (yellow). Double arrows indicate the reciprocal interactions between the two paths. Note: The activity between neurons of a same path is continuously circulated and fed back into the same network. In other words, the neural changes caused by any neuron in the given path is both fed forward to the next one in the hierarchy, and fed back to its original source. Such mechanisms consent both the broadcast of data, and the comparison of processual outcome between different areas. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
experiences. Brain imaging studies have additionally showed that, as an individual gains insight in own conscious experience, the neuromagnetic signals appear stronger, while the neuronal cluster becomes broader including frontal, parietal, temporal and occipital areas (Vaitl et al., 2005). At the light of existing evidence, but notwithstanding the need for further research, it appears that the suggestion about consciousness emerging from an endogenous feedback network, which allows for localized activation to become distributed knowledge across the brain, might indeed hold some promises. Fig. 5 shows the mechanism of recurrent interactions as hypothesized by the EFN theory. 3.4. On the function of consciousness There have previously been presented in this article some among the many explanations that have been offered to the function of consciousness. As it was mentioned, while some researchers argue that our consciousness is actually responsible for hampering our reasoning abilities (e.g. Bargh, 2002; Dijksterhuis & Nordgren, 2006; Lieberman, Gaunt, Gilbert, & Trope, 2002), other scientists (e.g. Crick & Koch, 1998; DeWall et al., 2008) highlight the role of consciousness in logical thinking, and in the general ability to process and select the most appropriate elements from the environment. Remarkably, most of such proposals, for as different from each other as they might be, seem to ascribe to consciousness a role in our cognitive mechanisms. Moreover, the spread negligence in differentiating between the terms ‘consciousness’ and ‘awareness’ appears to assume that the overt contents of our mental processes are in all the same as the contents of our consciousness.44 Unfortunately, by buying into such a conceptualization of consciousness which so precisely adheres with cognition, we might end up with an even more abstruse riddle: Why do we possess two systems which focus on the same tasks, namely stimuli processing and cognitive elaborations? In the attempt to address consciousness’ functionality, it will here be proposed that the phenomenon is only indirectly entangled in stimuli’s processing and manipulation, while it is directly involved in the optimization and integration of the mental mechanisms that process and manipulate such stimuli. In sum, while the specialized areas in our cognitive system are the recipients of bottom-up stimulation, our consciousness is responsible for modulating top-down resources. Accordingly, two main features will identify the role of consciousness: Integration and Speed. 3.4.1. Integration As it has been previously argued (see On the emergence of consciousness), consciousness supposedly arises from the binding together of populations of neurons which together identify a distinct and widespread neural network, namely the EFN. It has further been argued that the role of such network is to broadcast the occurrence of bottom-up activation and stimuli’s processing, consequently allowing us to brand ownership on their outcomes (e.g. thoughts, perceptions, decisions, etc.). It will be interesting to notice at this point that the binding between neural clusters across distal brain regions appears dis-
44 While this assumption could be sustained when we refer to access- and self-consciousness, it is less plausible to argue that, at any one time, our cognitive system would be explicitly processing stimuli to which we have not given direct attention, or stimuli which we cannot (or refuse to) give attention, or others which we cannot even describe. These three kinds of stimuli can otherwise be assumed to fall under the implicit contents of monitoring-, non-cognitive-, and phenomenal consciousness, respectively.
568
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
rupted in patients affected by schizophrenia (Tononi & Edelman, 1998). The feeling of being externally governed (‘‘alien control”), and the sense of estrangement from own acts, typical of this disorder of consciousness, could in fact reflect the patients’ inability to rely on the integrative function offered by the consciousness network, and therefore to distinguish between self-generated and externally triggered activity. In other words, schizophrenia might impair or abort the correct functioning of the EFN, while preserving a relative functionality of the patients’ cognitive abilities to detect and process stimuli. The result would therefore be the individual’s awareness of thoughts, perceptions and decisions, in absence of any consciousness of them being the result of own cognitive mechanisms. Patients’ self-reports, as those offered by Artaud, can testify for the puzzlement and helplessness experienced by schizophrenic patients, as their awareness of voice-hearing and endless thinking is not met by their experience of being the originators of such artifacts. The EFN theory proposes therefore that integration is an important function of consciousness, and it will be quoted in support Vaitl and colleagues’ definition for conditions such as schizophrenia: ‘‘[disorder of consciousness are] disorder of integration between the sensory systems of consciousness and the motor systems of thought” (2005, p. 112, own emphasis). 3.4.2. Speed Both philosophy and neuroscience have agreed on the immediacy of our conscious experiences. As I taste the dark liquid in my cup, the experience of coffee hits me at once: Perception and experience appear to be two synchronous features of a same act. The ‘‘what-it-is-like-for-me-to-drink coffee” is however based on the fact that I know the taste of coffee, that I have already established that I like it, and that I might associate it to some specific feelings, such as coziness, energy or relaxation. In other words, my experience of coffee is conveyed to me by the full cluster of data I have learned to associate to the beverage, and this broadly nuanced experience emerges as soon as the liquid reaches my lips. However, as I taste my coffee, I am not aware of the many cognitive elements that shape such experience, indeed it appears the case that an interval of time (ca. 200 ms)45 separates my coffee experience from my ability to cognitively access it. In other words, we seem able to rely on an independent mental mechanism that enables us to receive representations about experiences even before we have been able to cognitively appraise their nature. As demonstrated by Castiello and colleagues, such pre-cognitive neural activity can initiate responses or, as van Gaal, Ridderinkhof, Fahrenfort, Scholte, and Lamme (2008) clearly showed, exert inhibitory control over a prospective task or interrupt an on-going one.46 Back to my coffee experience, as the liquid touches my tongue its impact triggers a specific signal. Following the EFN theory, it is suggested that this pattern is immediately broadcast, and it represents the cue for a specific data-cluster in my memory, which rapidly reproduces the experiential effect I have learned to associate with coffee. In fact, it would be an undeniable expenditure of mental resources if my cognitive system were to trace step-by-step the elements that compound in my ‘coffee-experience’ every time I take a sip. It is therefore argued that phenomenal consciousness is the mechanism responsible for such rapid broadcasting of perceptions. The suggested purposes of such prompt activations are: (1) to allow a certain swiftness in responses and/or to prepare the organism toward the detection of a given stimulus,47 (2) to facilitate the triggering of automatic action patterns based on mnemonic schema, (3) to allow experiential richness and immediacy without unnecessarily loading our cognitive system, and (4) to exert a prompt deterrent against possibly threatening stimuli. In sum, it is argued that consciousness enables us to take advantage of a pre-cognitive short-cut, a sort of stimulus’ ‘speedy-line’, leading us to a specific kind of intrinsic pseudo48-knowledge about our percepts, which precedes their full-fledged cognitive assessment. The conceptualization of emotions could in this context be expressed as a pre-cognitive, perceptual label raised by the specific cluster of data emerged following the broadcasting of a given perception. The so-called gut-feelings or intuitions49 could indeed support such arguments. Volz and von Cramon (2006) suggested that people are generally able to detect familiar patterns in the rich flow of perceptions constantly impacting their senses. Such percepts, which occur in lack of awareness, evolve in what the researchers called, ‘‘a vague perception”, which cannot be explicitly described, nor cognitively accesses50 – a gut-feeling indeed. Volz and von Cramon further suggested that specific cues, ‘‘by means of spreading activation, activate a mnemonic network which integrates the entire stream of prior experiences that are all critically related to the crucial event” (p. 2077). Finally, in absence of a pre-existing mnemonic trace, due to the complete novelty of the stimulus, it is hereby hypothesized that a ‘second best’-match might be activated, and we might form a perception based on our experience of stimuli and/ or circumstances similar or reminiscent of the novel one. Were the stimulus completely foreign, it is speculated that it might 45
See Castiello et al. (1991). Combining masking and Go/No-Go paradigms, Van Gaal and colleagues showed that an unconscious stimulus can produce inhibitory control of responses. Electroencephalographic recordings showed that the unconscious stimuli would first of all determine an activation in occipital frontal cortex and successively in the prefrontal cortex (PFC). Interestingly, while the first neural response was identical whether or not the unconscious stimulus had behavioral implications, the PFC activation appeared to modulate the implementation of inhibition only in case of No-Go stimulus. The results were taken to show that ‘‘unconscious stimuli can influence whether a task will be performed or interrupted, and thus exert a form of cognitive control” (2008, p. 8053). 47 Súper et al. (2003), studying the visual mechanisms of stimulus detection, showed that monkeys’ V1 area had to be at a specific level of activation prior to stimulus presentation in order for that stimulus to be perceived. 48 See Note 22 for the meaning attributed to the suffix ‘‘pseudo”. 49 The term ‘intuition’ is here used in agreement with Parker’s (1990) definition, that is, ‘‘a preliminary perception of coherence (pattern, meaning, structure) that is at first not consciously represented” (p. 74). 50 Volz and von Cramon’s ‘‘vague perception” appears also to reflect the concept of ‘‘sensing” considered by Galpin, Underwood, and Chapman (2008). Furthermore, the presence of a pre-attentive sensing-mechanism, seemingly able to detect changes prior to the subject’s explicit awareness of their occurrence, could indeed accommodate the conceptualization of phenomenal consciousness proposed by the EFN theory. 46
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
569
either be non-cognitively categorized as of low relevance or, depending on the salience of its features, eventually be granted more appropriate cognitive evaluation. 3.5. EFN – summary It might be useful at this point to attempt a summarization of the theory of consciousness presented in this section. It has been proposed that, as stimuli impact our sensory systems, the neural changes produced by their detection and further processing determine subsequent activations along parallel neural paths (or neural clusters). The specificity of such neural clusters is assumed to reside in their orientation toward the internally generated activity produced by stimuli processing, rather than in the stimuli per se. Bound together by recurrent activity, the globality of such neural clusters forms a widespread neural network, which has been denominated Endogenous Feedback Network (EFN). It has further been argued that the emergent property of the activation patterns initialized along the EF network identifies the phenomenon of consciousness. The role of the EFN has been hypothesized to reside in the broadcasting of the occurrence of localized mental activations, allowing them in this way to become distributed knowledge across the brain. Such distributed knowledge enables us to reciprocally integrate the processing of data, to coordinate conception, execution and perception of acts, to increase the speed and efficiency of responses, and to optimize the allocation of our mental resources. The EFN can therefore be conceptualized as the global, internal feedback that the brain forwards to itself, where the outputs of the EFN are perceived by us in form of internal stimuli. Having advocated for a distinction between the various aspects of consciousness (phenomenal, monitoring, access, noncognitive and self-consciousness), a specific role and profile has been assumed behind each of them: (1) Phenomenal consciousness – It informs about the detection of a stimulus; As no cognitive processing has yet been activated, the stimulus lacks any descriptive profile; No reportability, nor cognitive access available. Purpose: To prepare the organism toward the processing of a given stimulus, to facilitate the triggering of automatisms, to enable the emergence of pseudo-percepts (sort of emotional flags) which can allow for more rapid and efficient responses. (2) Monitoring consciousness – It follows the neural activation correlating with a given automatism triggered by a specific stimulus; It employs reduced mental resources; Reportability and cognitive access enabled at the moment that our response receives attention. Purpose: To facilitate multitasking behavior by monitoring the correct performance of familiar sequences without placing additional load on our cognitive system. (3) Access-consciousness – It is mediated by attention, and it correlates with stimulus’ relevance, novelty and/or complexity; It employs extended mental resources; Reportability and cognitive access enabled. Purpose: To closely follow the processing of particularly complex and/or novel stimuli, and to consequently allow their precise evaluation. (4) Non-cognitive consciousness – It follows the detection and (limited) processing granted to those stimuli to which – as determined by their low relevance, or high distressfulness, or meaninglessness – no cognitive evaluation has been granted; No reportability, nor cognitive access enabled (stimuli can receive cognitive resonance only if they can link to a monitoring or access-consciousness trace). Purpose: It vouches for cognitive repêchage by offering stimuli an additional chance to exert an effect. It also functions as ‘cognitive closet’ for psychologically threatening stimuli which could otherwise hamper the brain’s cognitive functions. (5) Self-consciousness – It identifies the EFN pattern responsible for broadcasting the overall effect of the activations reported by phenomenal, monitoring-, access-, and non-cognitive consciousness; Reportability and cognitive access enabled. Purpose: To broadcast a historical, causal and multidimensional neural trace for self-generated acts enabling in this way the individual to distinguish them from the actions of others; To probe and evaluate the experiential outcome of our actions, thoughts, and decisions. On the whole, it has been proposed that the different routes leading toward monitoring-, access-, or non-cognitive consciousness are determined by the type of activation caused by the stimulus’ initial assessment. In other words, the phenomenal trace, indistinctively activated by any stimulus able to produce a neural change in our brain, is broadcast by the EFN. Such signal will consequently trigger the stimulus’ early assessment based on the eventual activation of associable, memory-stored, data-cluster. Emotions could in such sense reflect the perception of specific non-cognitive markers linked to the given data-cluster. Absence of a pre-existing representation will determine the decay of the trace or – depending on stimulus saliency – a default allocation of access-consciousness. Finally, the EFN theory has suggested that the factors determining the conscious experience of given stimuli above others are strictly dependent on the specific phenomenal trace triggered by each stimulus. The EFN’s model (Fig. 6) sketches the dynamic between the various aspects of consciousness. 4. Section 3 4.1. EFN – strengths In this final section it will be evaluated to which extents the EFN has fulfilled its intentions to: (a) support the composite profile of consciousness emerged from the alliance between the selected group of philosophical and neuroscientific theories
570
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Fig. 6. The model illustrates the emergence of the five different aspects of consciousness, and the relationship between the brain’s cognitive systems (blue pattern) and the EF network (yellow pattern). (a) As stimuli reach our sensory systems, they induce in the blue network neural changes which are immediately detected by the EFN pattern. At this stage, an early signal is broadcast in terms of phenomenal consciousness. (b) The phenomenal trace might in turn activate a specific pattern in monitoring, access- or non-cognitive consciousness, eventually leading to a specific response. Note: (1) the EFN patterns can affect the cognitive system by means of their own activation: Such signals translate in fact into internal stimuli, and allow for the content of our consciousness to reach cognition; (2) Traces in non-cognitive consciousness can either emerge into awareness (if receiving specific priming), or they might indirectly affect behavior by triggering a response (e.g. automatism, emotion), or by determining attentional allocation (red dotted lines). (c) Any activation of the cognitive system (caused by the externally generated stimulus, and/or by the EFN feedback) is meanwhile producing continuous neural changes which are being synchronously detected and propagated by the EFN. The signals generated by the EFN, and those determined by the cognitive processing converge respectively in the streams of self-consciousness and self-awareness. Note: The minimal time interval that separates the broadcast of each activation-‘snapshots’, and the fact that the two patterns synchronically follow each other in a loop of constant, reciprocal activations, determine the precise integration of the perceptions we acquire at any one time, and the experience of a unified and coherent flow of consciousness. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
of consciousness (see Table 2) and (b) to compensate for the explanatory gaps that such theories had left behind (see General considerations, Section 2). Table 3 places the EFN at the light of existing knowledge.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
571
4.2. EFN – integrating new theoretical addenda to existing knowledge The theories of consciousness reviewed in the 1st section of this article had left a number of questions unanswered, showing in such way specific limits of their theoretical span, or perhaps some flaws in their ground assumptions. The EFN framework has attempted to fill such explanatory gaps. 4.2.1. On the emergence of consciousness of a stimulus but not of others The EFN theory has proposed that the mere exposure to a stimulus is sufficient to determine the broadcasting of the neural changes caused by its detection, and by the early processual stages.51 Furthermore, the pattern triggered by the specific input has been assumed to consequently determine the emergence of a phenomenal trace, granting the stimulus the possibility to activate a specific data-cluster stored in our memory. On the basis of: (1) such eventual mnemonic match, (2) of the consequent (albeit rudimentary) attribution of meaning and relevance, (3) of the perceptual saliency of the stimulus, and (4) of the contextual availability of the individual’s cognitive resources, a specific stimulus might be acknowledged above others. 4.2.2. On the emergence and function of phenomenal consciousness As mentioned above, phenomenal consciousness is supposed to identify the broadcasting of the neural changes determined by a stimulus’ detection and early processual stage.52 The function of phenomenal consciousness is therefore assumed to reside in its ability to convey a most rapid feedback about the possible meaning(s) of the stimulus, based on the specific memory data-cluster that such broadcasting might have activated. Its evolutionary value will consequently consist in granting to the organism the possibility to react, or to be predisposed toward, or eventually to deny further expenditure of cognitive resources to stimuli, based on the specific pre-cognitive labels they have generated. 4.2.3. On the factor(s) that allow a given emotion to become conscious, and on the reasons why we should experience emotions at all In this paper it has been suggested that emotions, mediated by the emergence of phenomenal consciousness, represent a sort of non-cognitive flags assigned to those data-clusters we have acquired familiarity with, or which meanings have particular relevance to us. Emotions can consequently be conceptualized as mental short-cuts to specific percepts without the need to activate the full data-clusters that have determined them and that cognitively justify them.53 Upon re-exposure to a known stimulus, and to the consequent activation of an associable cognitive and experiential cluster, the emotional ‘flag’ will in fact lead us concisely and directly to the response we have learned to associate to the percept. Having suggested that all mental activity produces an effect along the EFN, and that such effect is always broadcast and therefore allowed to trigger associable matches, the EFN theory can agree with Thagard and Aubie (2008) in their proposal that emotions likely accompany all mental acts. However, it will be argued that emotions emerge based on the relevance of the stimulus to which they link. In other words, given that an input, on the basis of its early evaluation, has been categorized as irrelevant or possessing low priority status, the emotions attached to it would likely not acquire perceptual definition. More so, having assumed that emotions are the experiential expression of an existing data-patch, neutral stimuli (that is, lacking noteworthy significance to the individual) will likely not determine the emergence of any emotion at all, since their perception wont meet any relevant memory cluster. 4.2.4. On a system able to simultaneously present the distinct representation of each stimulus appearing at one time, while still conferring the unitary perception of the overall experience It has been proposed in this article that the EFN reflects all mental activity occurring at any one time anywhere in the brain. This assumption implies therefore the broadcasting of the neural changes caused by each stimulus singularly impacting our sensory and processing system. However, given the suggestion that such global broadcast is constantly active, and that it therefore functions as a full rounded loop, the activation-units merge and integrate with each other on a continuum. The single frames of mental life will in such way acquire the perceptual and distinctive flavor of a unified flow of consciousness. 4.3. EFN – empirical evidence The fact that many of the EFN’s theoretical elements can be supported by the tenets of a relevant number of existing theories (see Table 3), confers double strength to the framework. In fact, not only the EFN can acquire theoretical plausibility by showing compatibility with the thoughts of such notable group of scientists, but it can also rely on the bulk of supporting evidence that those previous accounts have generated.
51
Such early activations appear to occur in absence of awareness (see Escera et al., 2003). A very interesting study carried on by Lamy et al. (2008) greatly supports the occurrence of spread processual activity preceding cognitive awareness of the stimulus. 53 Note however that some emotions might not be supported by any actual cognitive cluster, as they might have instead been acquired through conditioning and reinforcement, e.g. phobias triggered by harmless insects. 52
572
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
The sample of evidence already offered in support of the EFN theory and presented in the previous section might have hopefully paved the road at least to the possibility of the presence of distinct neural paths respectively modulating phenomenal, monitoring, access, non-cognitive and self-consciousness. Finally, it will now be recalled the work of a few other relevant researchers to support the proposal that such distinct neural patterns collectively come to shape a specialized structure: The Endogenous Feedback Network. 4.3.1. Anaesthesia: impairing the feedback Alkire, Hudetz, and Tononi (2008) aimed at assessing the extents to which anaesthetics can be reliably trusted to determine unconsciousness. Based on the many cases of patients waking up able to recall events that had taken place in the operating room during surgery, the authors stressed that anaesthesia does not necessarily imply absence of consciousness as a whole. Alkire and colleagues observed that anaesthesia is associated with deactivation of mesial parietal cortex, posterior cuneus and posterior cingulate cortex, which are all structures likely involved in global monitoring and in other functions related to the self54 (Gyulai et al. 1996; Vogeley et al., 2004). However, the most consistent sign of induced loss of consciousness appeared to be a decrease in thalamic metabolism and blood flow: Intriguingly, thalamic activity did not decrease with the administration of anaesthetics (Alkire et al., 2008). The effects of the drug appeared rather to play indirectly on the thalamus by altering the cerebral activity in the cortical areas to it associated. Vahle-Hinz, Detsch, Siemers, and Kochs (2007) have congruently demonstrated that the thalamus of animals to which the cortex had been removed appeared unaffected, both metabolically and electrophysiologically, by the administration of anaesthetics. The conclusion reached by Alkire and colleagues in their study had been that the thalamus must be understood as a readout of global cortical activity, and as a modulator of functional integration between distal brain areas. The authors therefore suggested that anaesthesia might not produce its effects so much in terms of inhibiting the neuronal activity in cortical structures, but rather in disabling the thalamus’ ability to integrate information, and in slowing/inhibiting neural responses provoking in such way a loss of synchronization among distal areas.55 It will be interesting to note at this point that the neurophysiological mark of anaesthesia is a drop in EEG coherence in the gamma-frequency range between opposite cortical hemispheres, and between frontal and caudal areas (Alkire et al., 2008). As it has already been mentioned when discussing the emergence of consciousness, Tononi and Edelman (1998) have argued that conscious experience emerges as the separate neuronal groups in the thalamo-cortical system (taken to identify different mental states) bind together following recurrent interactions. Through the work of Llinás and Paré (1991), such neural binding appeared to be sparked namely by gamma oscillations at 40 Hz. In sum, the evidence reported above converges in the proposal that the loss of consciousness caused by anaesthesia is determined by the disruption of the brain’s ability to integrate information, and to synchronize the activities in distal areas of the brain. Given therefore that anaesthetics seem to target the specific neural ‘‘entity” responsible for such tasks, the EFN suggestion about the presence of a neural network responsible for data integration and processual coordination appears to acquire renewed plausibility. 4.3.2. Recurrent neural activity: building the network The EFN theory likely finds its most solid support in the work of Victor Lamme, who strongly argued that the only way to demystify consciousness is by viewing it separate from cognition (Lamme, 2006a). Resting his arguments on a massive bulk of physiological evidence,56 (Lamme, Supér, & Spekreijse, 1998; Lamme, 2000, 2001, 2003, 2004, 2006b) has demonstrated the presence of two kinds of neural activation – feedforward sweep and recurrent processing – which follow the presentation of a stimulus. The feedforward sweep (FFS), definable as the unconscious, earliest activation of cells in successive areas of the cortical hierarchy, travels at remarkable speed. In fact, following stimulus’ presentation, a 10 ms latency is sufficient for the information to move from one level to the next, that is, cells in V1 activate at 35 ms, in dorsal areas MT and MST at 45 ms, and frontal eyes’ fields cells at 55 ms (Lamme, 2004). These short intervals indicate therefore that only one spike has been fired by a cortical neuron before the next level has been activated.57 However, despite their speed, these early spikes can convey information such as object’s orientation, shape, color, faces’ detection, and motion perception (Lamme, 2001). In other words, FFS is able to bear all the necessary information to allow the individual to perform basic categorizations (such as object/non-object, face/non-face), and even to trigger automatic responses (Lamme, 2001, p. 214). Nevertheless, according to Lamme (2006a), the signals sent through the FFS do not have the prerequisite necessary to generate conscious experience. Consciousness will rather
54 Subjects, who had received doses of a specific kind of anaesthetic (nitrous oxide) – which appeared to selectively deactivate the posterior mesial cortex – reported upon awakening having experienced gradual, dreamlike feelings with depersonalization and out-of body experiences, but they would not consider themselves as having lost consciousness altogether (Gyulai, Firestone, Mintun, & Winter 1996). The authors correlated the graduality of such experiences to the progressive switching off of the mesial cortex. 55 Computer simulation of the effects of disruption showed in fact that it is sufficient to interrupt a relatively limited number of connections in order to fully demolish overall communication within the network (Steyn-Ross, Steyn-Ross, & Sleigh 2001). 56 Lamme’s work on the mechanisms behind visual processing has mostly been concerned with experiments on the macaque’s visual system. 57 Estimation reached given a maximal rate of neural firing of 100 Hz (Tovee, 1994, in Lamme (2001)).
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
573
emerge only once information from higher areas – by means of recurrent processing (RP) – are made available to lower ones, and allowed to compare the respective interpretations of the stimulus.58 With Lamme’s words, ‘‘Initial responses reflect sensory processing by the feedforward sweep, whereas responses at longer latencies correlate with more cognitive or behavioral aspects on sensory processing59,60 (2001, p. 217). Lamme’s findings, and partially the interpretations he offers of them,61 appear to support the main tenets of the EFN’s theory. The early onset and speed of the FFS, its ability to convey information valuable enough to trigger an automatic response and/or to endow the individual with a representation (albeit raw) of the incoming stimulus, are features in perfect adherence with the EFN’s profile of phenomenal consciousness, and with its conceptualization as the gateway to other forms of consciousness. Furthermore, Lamme’s suggestion that consciousness is the result of recurrent interactions between groups of neurons, which allow information from higher areas to be fed back to lower ones, is in accord with the EFN’s proposal of a neural broadcasting and endogenous feedback taking place within and between all processing areas in the brain. Lastly, the occurrence hypothesized by Lamme of a comparison of processual outcomes between higher and lower brain areas, is only within a linguistic stone’s throw from the EFN’s conceptualization of a neural network, which purpose is to reciprocally integrate processual information of various nature. Fig. 7 summarizes the strengths of the EFN, and it further indicates its theoretical addenda. 4.4. EFN – limits As Crick and Koch (2003, p. 119) anticipated, a good framework for consciousness, while expected to sound reasonably plausible relatively to available empirical evidence, it is also unlikely correct in all its details. The EFN wont escape such prophecy, as it is certainly not immune from flaws. First of all, it should be pointed the judgmental finger against the very ground that holds the EFN framework, that is the group of existing theories that appears to confirm so many of the new theory’s tenets. It could in fact be objected that the very selection of the accounts presented in Section 1 – besides certainly being quantitatively restricted – has been biased by the effort to find evidence that would support the composite profile of the EFN. In other words, by supporting and being supported by a chosen group of existing theories, the proposed framework could represent a sort of self-fulfilling prophecy. Furthermore, the strength drawn from the reviewed theories of consciousness could potentially turn into a double-edged knife. In fact, criticism built against any of the empirical elements borrowed to advocate the EFN is likely going to impact on the tissue of the newly proposed framework as well. It will therefore be underlined the necessity to gather new and ad hoc evidence that might reinforce the EFN’s theoretical independence, and help to counterbalance for the above-mentioned pitfalls. Another criticism toward the EFN could then be that the bulk of evidence presented to support the proposal of distinct neural patterns respectively defining consciousness and cognition could be misleading. In fact, it cannot be discarded the possibility that the parallel neural activity identified in some of the studies reported previously in this article, could truly point to a correlative activation between patterns, rather than to a causal one. In other words, altogether different mechanisms, linking directly or indirectly to cognition, could be behind the emergence of consciousness. Activity in such unrevealed mechanisms could then determine the pattern of neural activation (that is, what has been called ‘EFN’) supposedly running parallel to our cognitive mechanisms. Germane to this possibility, it could therefore be hypothesized that consciousness might itself be the cause and not the product of a parallel neural networks. Unfortunately, the ability to identify and follow a distinct neural path among the intricate labyrinth of connections unfolding in our brains, and the possibility to separate – both in space and time – its specific activation with respect to others, creates non-negligible difficulties to our present technologies. Nevertheless, while waiting for technical science to catch up with our needs, properly designed experiments – by empirically strengthening cause and
58 Among the rich empirical evidence that he put forward, Lamme described the electrophysiological profile of subjects during backward masking trials, arguing that feedforward activation is not suppressed by the mask, while instead the recurrent activity is. In fact, the ms interval between the 1st stimulus and the mask is such that by the time the feedback sweep has reached again V1, it will be met by the feedforward sweep determined by the 2nd stimulus. The mismatch will cause the abrogation of the 1st input, and therefore the subject inability to report it. Lamme has also found that FFS is still detectable in anaesthetized animals, while recurrent processing are reduced or aborted (Lamme et al. 98). In advocating the role of recurring activity, Lamme further underlined that it respected the principle of Hebb’s rule (1949), which states that pre- and post-synaptic neurons have to be simultaneously active in order to facilitate synaptic plasticity processes, and to promote changes in the brain. Following Watanabe’s (2001) argument, the FFS activity per se would therefore produce only short-lived effects. 59 Expressing such neural dynamics in temporal parameters, we will have the activation of the feedforward sweep starting at stimulus onset and terminating 120 ms later, having meanwhile reached all processing areas in the brain. At such point, the recurrent activity will rouse, leading eventually to attentional changes which won’t be detectable before 200ms after stimulus onset (Lamme, 2001). 60 Lamme’s argument is also supported by Zhang, Riehle, Requin, and Kornblum (1997), who reported that early response in the motor cortex identifies the sensory detection of the stimulus, while response at intermediate latency appears to code for sensory-motor mapping, and even later responses are linked to motor commands. 61 Disagreement between Lamme’s interpretations and the EFN’s suggestions resides for example in Lamme’s assumption about the role of the FFS which he considers extraneous to consciousness’ mechanisms, while it is identified within the EFN frames as phenomenal consciousness. Furthermore, Lamme suggests that RP determines the emergence of consciousness from the same neural patterns engaged in cognitive processing. To our knowledge, at no point in his work Lamme hypothesized the presence of a distinct neural path strictly modulating consciousness, as instead it is proposed by the EFN theory.
574
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Fig. 7. The drawing shows the compatibility of the EFN theory with the integrated profile of consciousness emerged by linking the points of agreement between neuroscientific and philosophical theories of consciousness considered in this article (see Table 2). The blue boxes sum up the respective main tenets (the arrows link each of these to the neuropsychological and/or philosophical account/author that has promoted them). Furthermore, the EFN appears able to supplement for the four points of enquiry left unanswered by all other theories reviewed ahead (the green squares indicate such theoretical addenda). The fragmented line symbolises the intra- and inter-disciplinary compatibility between the different accounts, giving shape to a composite, but remarkably unified, profile of consciousness revolving around the EFN. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
effect relationships between brain activations – might help us to shrink the portfolio of possibilities lying in front of us. Lastly, it cannot be escaped the recognition that the EFN framework still has not told us how – concretely speaking – our conscious experience can explode in all its subjective richness from the perhaps less phantasmagorical chemistry between neurons. On the other hand, we might have to consider the possibility that the ‘chemical conversations’ taking place in our brains might just not be translatable into human language, in which case our hopes to grasp how chemistry can unfold into consciousness might remain disappointed. Nevertheless, we trust that an alternative path exists, able to bring us equally close to the core of the mystery. It could consist of efforts to understand the rules and the mechanisms of the neural language that determines consciousness. For let alone we might perhaps never grasp how ‘the water of objective processes turns into the wine of subjective experience’, just to borrow Chalmer’s (1996) captivating metaphor. But at least we should be able to get a good grip on what determines the change, why it occurs, and where the metamorphosis begins.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
575
Appendix A Box 1: EFN Theory – Glossary. Table 1: Philosophy of consciousness. Table 2: Integrated profile of consciousness. Table 3: Degree of compatibility between EFN and other existing theories of consciousness.
Table 1 The schema offers a succinct overview of the main assumptions supporting a selected group of philosophical approaches to consciousness. Philosophy of consciousness Theories
Tenets
Mysterianism
Although the mystery of consciousness could be demystified, we just do not possess the sufficient mental abilities to do so (cognitive closure) Our mental inadequacy is determined by the fact that, while our brains are designed to represent objective facts possessing a feature of spatiality, consciousness lacks such dimension (it could perhaps be located in the head, but we could not say anything about its size, shape, etc.) We may understand complex concepts only if these can be broken down into smaller and simpler levels of analysis
Dualism
The alleged distinction between the two foremost components of the world, physical and thinking stuff (Lat. res) has been taken as explanation for the mysterious nature of consciousness Our brains cannot understand the mystery of consciousness because their physical nature belongs to an altogether different dimension from the non-physicality of experience. However, thought-based research might allow us to explain consciousness in physical terms, but only in the sense that we might be able to learn what causes consciousness: How and why would remain for us mysteries
Representationalism
Phenomenal experiences may be reduced to the physical mechanisms of our brains. However, since consciousness is assumed to be externally based, the brain is considered merely as the screen on which the experience is played The feeling of an experience is determined by the representation of the percepts stored in our brains where they are coded in terms of their shape, color, form, etc. Following such tenet, the theory draws a tight and implicit bond between conscious experience and our cognitive mechanisms Phenomenal experience can reach us just by virtue of stimulus exposure: Introspection is not a determinant for conscious experience
HOT theories
Reflecting the brain’s hierarchical structure, thoughts cross processual levels of increased sophistication as they reach higher cortical layers An unconscious mental state (1st order thought) becomes conscious in the moment we make it the object of a second thought An inner sense organ in our brains is suggested to be responsible for the scanning of the outputs of 1st order thoughts (unconscious perception), and for the generation of the phenomenological aspects of the experience (Inner Sense theory) Consciousness is the result of global, meta-representation mechanisms
Brentanianism
The existence of unconscious mental activity is rejected: Consciousness is an intrinsic property of all mental states As all mental states are directed first toward the primary object (implicit self-awareness, triggered by simple exposure to stimulus) and secondarily toward themselves, conscious experience is always a reflexive phenomenon All mental processes, even those that we have not introspected, contain a primitive degree of self-awareness. All mental acts possess the dual ability to initiate both conceptual and not conceptual processes The feeling of any experience depends on the kind of judgement we have formed about the percepts that compound it
Phenomenology
We are not able to appraise the world directly, but rather only as it appears to us Since all conscious thoughts are experiences and experience is determined by the individual’s encounters with the world, consciousness is therefore never self-enclosed To know and to feel are intentional qualities of a same mental act: Self-awareness is an intrinsic phenomenon of any experience Experience is determined by our perceptions and our body is the very vehicle of perception: As perceptions are not localized, consciousness appears to be the product of a distributed system
576
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
Table 2 The table offers a composite profile of consciousness as it emerges by linking the points of respective strength and mutual compatibility between the neuroscientific theories of consciousness considered in this section and the philosophical accounts presented in Table 1. Integrated profile of consciousness (C) Features of intra-disciplinary compatibility NEUROSCIENTIFIC THEORIES
Inter-disciplinary agreement between neuroscience and PHILOSOPHICAL APPROACHESa
C emerges from the same mental act of processing a Phenomenology; Representationalism; stimulus (Thagard & Aubie, 2008) Brentanianism
C is the result of a distributed system involving many brain areas (Thagard & Aubie, 2008)
HOT theory; Phenomenology Representationalism
C is the mental act of processing own representations (Kriegel, 2007)
Actualist/Dualistic HO theories Representationalism; Brentanianism; Phenomenology
C emerges from mental broadcasting of stimulus processing (Baars, 1988)
HOT theory; Phenomenology; Representationalism; Brentanianism
C is independent from attention (Crick & Koch, 1998; Damasio, 1999)
Actualist/Dualistic HO theories Phenomenology; Brentanianism
C is linked to working memory (Baars, 1988)
Representationalism; Phenomenology; Brentanianism
C is admissibly emergent also in non-human animals (Crick & Koch, 1998) C comprises two distinct processual modes: Cognitive and non-cognitive (Crick & Koch, 1998)
Representationalism; Brentanianism Actualist/Dualistic HO theories; Brentanianism; Phenomenology
C is the optimizer of stimulus selection (Crick & Representationalism Koch, 1998) C can be initialized by mere exposure to stimulus Representationalism Brentanianism; (Damasio, 1999) Phenomenology C comprises four distinct kinds of experience, respectively defining: Monitoring-, access-, self-, and phenomenal consciousness (Block, 1995)
Features’ strength
Since consciousness’ emergence is understood as intrinsic in stimuli processing, immediacy and intentionality of conscious experience can be explained Resilience to variety of brain damages can be explained; the involvement of many brain structures can also explain our ability to experience conscious states differing per content, quantity and quality Access-, monitoring-, and self-consciousness can be explained: Their contents might reflect in fact our knowledge about own mental representations, thoughts and feelings The broadcasting across many different areas could account for a different quality of conscious state depending on the processual stage of the stimulusb Consciousness during sleep, as well as in other instances (e.g. veg.states) is admissible; Occurrences in which consciousness about stimuli precedes allocation of attention is explained Proposal can explain consciousness of stimulus’ absence (as based on individual’s expectations of stimulus); It can also explain the subjective character of experiencec Non-human consciousness research is admissible Proposal can explain cognition in absence of consciousness and vice versa: The two mechanisms, while supposedly intrinsic in the processing of each stimulus, are assumed distinct from each other and therefore possibly also separable Proposal can explain consciousness in terms of its functionality, conferring to it an evolutionary value. No need to assume pre-cognitive selection of stimulus The separate investigation of distinct aspects of consciousness might lead to a better understanding of the phenomenon as a whole (Rosenthal, 2005)
a
See Table 1. This feature is based on our assumption that phenomenal, access, monitoring and self-consciousness reflect specific stages of stimulus processing which are qualitatively and quantitatively different from each other. c It is implicit in this statement our suggestion that the subjectivity of our experiences is given by the fact that they are affected by our stored memories of meanings and previous encounters we have individually accumulated about the specific object of experience. b
Table 3 The table shows the degree of compatibility of the EFN theory with respect to the composite profile of consciousness presented in Table 2. Neuroscientific feature
EFN theory
C emerges from the same mental act of processing a stimulus (Thagard & Aubie, 2008)
Accordingly, the EFN suggests consciousness as the emergent property of a neural network activated by the neural changes generated by the processing of stimuli The EFN framework has compatibly suggested that consciousness reflects the activity of specific neural clusters distributed across the brain. The binding together of such clusters determines the broadcast of localized activations, and the emergence of distributed knowledge This assumption is perfectly compatible with the EFN framework (see point above) EFN theory fully agrees with this assumption. It further suggests that the distinctions between phenomenal, monitoring, access, non-cognitive and self-consciousness are rendered by the fact that they respectively reflect specific kinds of knowledge about given representationsa
C is the result of a distributed system involving many brain areas (Thagard & Aubie, 2008)
C emerges from mental broadcasting of stimulus processing. (Baars, 1988) C is the mental act of processing own representations (Kriegel, 2007)
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
577
Table 3 (continued) Neuroscientific feature
EFN theory
C is independent from attention. (Crick & Koch, 1998; Damasio, 1999)
This proposal is in agreement with the EFN suggestion that consciousness is distinct from wakefulness and from awareness. Attentional allocation is assumed to eventually follow detection and early processingb This assumption integrates point above, and it is fully compatible with EFN theory This proposal, supported by EFN theory, can account for our ability to gain consciousness about stimuli’s absence based on our expectationsc. Furthermore, the influence of WM is determinant for the subjective flavor of our percepts, which is assumed to be based on our history of encounters with the given object, and with the meaning we have learned to associate to it Having proposed that consciousness reflects the workings of the mental machinery by broadcasting the brain’s activations occurring at any given point in time, such occurrence could be admissible also in non-human animalsd. In fact, the EFN advocates that consciousness is not subordinate to the ability to report about own mental states nor – for that matter – to the subject’s awareness of them Compatibly, the EFN has further elaborated on this distinction, proposing non-cognitive consciousness as an integrative element of the phenomenon in its whole The EFN, in agreement with such proposal, argues that consciousness, by broadcasting the activations triggered by a stimulus even before its full cognitive appraisal has taken place, facilitates a hierarchical distribution of mental resources based on the saliency and relevance of the detected stimulus Block’s distinctions are proposed to reflect distinct phases of stimuli’s processing, and to identify the outputs of neuronal clusters determined by specific kinds of brain’s activity Next to Block’s four categories, the EFN introduced non-cognitive consciousness, able to account for those instances where apparently unperceived stimuli are still able to determine an experience (e.g. the gutfeeling) and/or to be cognitively accessede at later points in time
C can be initialized by mere exposure to stimulus (Damasio, 1999) C is linked to working memory (Baars, 1988)
C is admissibly emergent also in non-human animals (Crick & Koch, 1998)
C comprises two distinct processual modes: Cognitive and noncognitive (Crick & Koch, 1998) C is the optimizer of stimulus selection (Crick & Koch, 1998)
C comprises four distinct kinds of experience, respectively defining: Monitoring-, access-, self-, and phenomenal consciousness (Block, 1995)
a
A representation is intended as a ‘thought-unity’ containing all the stimuli contributing to a specific and meaningful percept. This assumption is also compatible with the Theory of Visual Attention (Bundesen, 1990). The TVA argues in fact that attention and selection do not occur prior to recognition, but rather through the mediation of memory, which is itself involved in the assessment of the stimulus. c Absence of stimuli could be detected due to lack of expected mental activation (e.g. entering a restaurant and realizing the absence of sounds as silence produces a mis-match with the memory-stored ‘public place’-schema), or due to inhibitory/excitatory effects caused by the cessation of the stimulus. d Speculatively speaking, one of the differences between human and non-human consciousness might therefore consist in their respective ability to rely on adequate brain sophistication to determine different levels of consciousness. e As per effect of the Cognitive Interview (Geiselman’s et al., 1985). b
Box 1. EFN theory – Glossary. Consciousness: It is a unitarian phenomenon comprising 5 distinct processual modes: phenomenal, access-, monitoring-, non-cognitive, and self-consciousness. (See Section 2 for an account of each mode). It indicates the broad neural network able to produce any of the above-mentioned 5 categories of percept. It broadcasts the neural changes determined by the perception, processing and manipulation of stimuli (both internal and external). Conscious processes: These are the processes of the consciousness network, and therefore not the ’processes of which a subject is aware’. Definition of a subject as ‘conscious’: It refers to the subject’s ability to rely on the brain’s capacity to trigger the neural mechanism(s) leading to at least 1 of the above mentioned 5 consciousness-modes. To be conscious of a stimulus: It is understood as ‘‘to initialize conscious processes associated to the detection and processing of a stimulus”, let it be whether such processes will be at phenomenal, access, monitoring-, non-cognitive, and/or self-consciousness level. To be conscious of ‘‘X” is therefore not intended as ‘‘to be aware of X”. Cognition: This term is used to indicate the broad range of mechanisms involved in the perception, processing and manipulation of stimuli (both internal and external). Awareness: This phenomenon is understood as possessing a strictly cognitive nature (in the sense of taking stimuli as its direct objects), and it is therefore not necessarily bound to consciousness. In the terms of the EFN theory, we can therefore be conscious of ‘‘X” even without being aware of it. Unconsciousness: The term is taken as ‘‘consciousness without awareness”. Having bound awareness to cognition, the EFN theory introduces the term ‘non-cognitive consciousness’ (see Section 2).
578
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
References Alkire, M. T., Hudetz, A. G., & Tononi, G. (2008). Consciousness and anesthesia. Science, 5903(322), 876–880. Atkinson, A. P., Thomas, M. S. C., & Cleeremans, A. (2000). Consciousness: Mapping the theoretical landscape in trends. Cognitive Sciences, 4(10), 372–382. Baars, B. J., & Gage, N. M. (Eds.). (2007). Cognition, brain, and consciousness. Elsevier Ltd.. ISBN:9780123736772. Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge: Cambridge University Press. Baars, B. J., & Franklin, S. (2007). An architectural model of conscious and unconscious brain functions: Global workspace theory and IDA. Neural Networks, 20, 955–961. Bargh, J. A. (2002). Losing consciousness: Automatic influences on consumer judgement behavior and motivation. Journal of Consumer Research, 29(2), 281–285. Bering, J. M., & Shackelford, T. K. (2004). The causal role of consciousness: A conceptual addendum to human evolutionary psychology. Review of General Psychology, 8(4), 227–248. Block, N. (1991). Consciousness and accessibility. Behavioral and Brain Science, 13, 596–598. Block, N. (1995). On a confusion about a function of consciousness. Behavioral and Brain Sciences, 18, 227–247. Block, N. (2007). Consciousness, accessibility, and the mesh between psychology and neuroscience. Behavioral and Brain Sciences, 30, 481–548. Bosinelli, M. (1995). Mind and consciousness during sleep. Behavioral Brain Research, 69(1–2), 195–201. Bosse, T., Jonker, C. M., & Treur, J. (2008). Formalisation of Damasio’s theory of emotion, feeling and core consciousness. Consciousness and Cognition, 17, 94–113. Braisby, N. (2002). Consciousness. In T. Cooper & I. Roth (Eds.), Challenging psychological issues. Milton Keynes: The Open University. ISBN:074925355X. Bundesen, C. (1990). A theory of visual attention. Psychol Reviews, 97(4), 523–547. Cahill, C., & Frith, C. D. (1996). A cognitive basis for the signs and symptoms of schizophrenia. In C. Pantelis, H. E. Nelson, & T. R. E. Barnes (Eds.), Schizophrenia: A neuropsychological perspective (pp. 373–395). New York: Wiley and Sons. Campbell, K. B. (2000). Introduction to the special section: Information processing during sleep onset and sleep. Canadian Journal of Experimental Psychology, 54(4), 209–218. Castiello, U., Paulignan, Y., & Jeannerod, M. (1991). Temporal dissociation of motor responses and subjective awareness. A study in normal subjects. Brain, 114(Pt. 6), 2639–2655. Chalmers, D. J. (1996). Facing up to the problem of consciousness. In Proceedings: Toward a scientific basis for consciousness. MIT Press. Cohen, G. (2003). Everyday memory. In G. Cohen, G. Kiss, & M. Le Voi (Eds.). Memory current issues. Open University Press. ISBN:0335190790. Coleman, M. R. (2007). Do vegetative patients retain aspects of language comprehension? Evidence from fMRI. Brain, 130(Pt. 10), 2494–2507. Crick, F. (1984). Function of the thalamic reticular complex: The searchlight hypothesis. Proceedings of the National Academy of Sciences of the United States of America, 81, 4586–4590. Crick, F., & Koch, C. (1998). Consciousness and neuroscience. Cerebral Cortex, 8, 97–107. Crick, F., & Koch, C. (2003). A framework for consciousness. Neuroscience, 6(2), 119–126. Dalton, P., & Lavie, N. (2004). Auditory attentional capture: Effects of singleton distractor sounds. Journal of Experimental Psychology: Human Perception and Performance, 30(1), 180–193. Damasio, A. (1999). The feeling of what happens: Body,emotions and the making of consciousness, Harcourt Brace, ISB 0151003696. David, A. (1999). Auditory hallucinations: Phenomenology, neuropsychology, and neuroimaging update. Ada Psychiatrica Scandinavica, 99(Suppl. 395), 95–104. De Gennaro, L., Ferrara, M., Cristiani, R., Curcio, G., Martiradonna, V., & Bertini, M. (2003). Alexithymia and dream recall upon spontaneous morning awakening. Psychosomatic Medicine, 65(2), 301–306. Dehaene, S., & Nacchade, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and workspace framework. Cognition, 79, 1–37. Delacour, J. (1997). Object perception and recognition: A model for the scientific study of consciousness. Theory & Psychology, 7, 257–262. DeWall, C. N., Baumeister, R. F., & Masicampo, E. J. (2008). Evidence that logical reasoning depends on conscious processing. Consciousness and Cognition, 17, 628–645. Dijksterhuis, L. P., & Nordgren, L. F. (2006). A theory of unconscious thought. Perspectives on Psychological Science, 1(2), 95–109. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data. MIT Press. ISBN:0262050471. Escera, C., Alho, K., Winkler, I., & Näätänen, R. (1998). Neural mechanisms of involuntary attention to acoustic novelty and change. Journal of Cognitive Neuroscience, 10(5), 590–604. Escera, C., Yago, E., Corral, M. J., Corbera, S., & Nuñez, M. I. (2003). Attention capture by auditory significant stimuli: Semantic analysis follows attention switching. European Journal of Neuroscience, 18, 2408–2412. Frith, CD. (1979). Consciousness, information processing, and schizophrenia. British Journal of Psychiatry, 134, 225–235. Frith, C. D. (1992). The cognitive neuropsychology of schizophrenia. Hove, UK: Lawrence Erlbaum. Galpin, A., Underwood, G., & Chapman, P. (2008). Sensing without seeing in comparative visual search. Consciousness and Cognition, 17, 658–673. Geiselman, R. E., Fisher, R. P., MacKinnon, D. P., & Holland, H. L. (1985). Eyewitness memory enhancement in the police interview: Cognitive retrieval mnemonics versus hypnosis. Journal of Applied Psychology, 70(2), 401–412. Georgieff, N., & Jeannerod, M. (1998). Beyond consciousness of external reality: A ‘‘Who” system for consciousness of action and self-consciousness. Consciousness and Cognition, 7, 465–477. Giacino, J. T., & Malone, R. (2008). The vegetative and minimally conscious state. In G. B. Young & E. F. M. Wijdicks (Eds.). Disorders of consciousness. Elsevier B.V.. ISBN:978044518958. Gyulai, F. E., Firestone, L. L., Mintun, M. A., & Winter, P. M. (1996). In vivo imaging of human limbic responses to nitrous oxide inhalation. Anesthesia and Analgesia, 83(2), 291–298. Heywood, C. A., Kentridge, R. W., & Cowey, A. (1998). Cortical color blindness is not ‘‘blindsight for color”. Consciousness and Cognition, 7(3), 410–423. Hobson, P. R., Chidambi, G., Lee, A., & Meyer, J. (2006). Foundations for self-awareness: An exploration through autism. Monographs of the Society for Research in Child Development, 71(2), vii–166. Kentridge, R. W., Heywood, C. A., & Weiskrantz, L. (2007). Color contrast processing in human striate cortex. Proceedings of the National Academy of Science of the United States of America, 104(38), 15129–15131. Kircher, T. T. J., & Leube, D. T. (2003). Self-consciousness, self-agency and schizophrenia. Consciousness and Cognition, 104(38), 15129–15131. Koch, C., & Tsuchiya, N. (2006). Attention and consciousness: Two distinct brain processes. Trends in Cognitive Sciences, 11(1), 16–22. Kouider, S., & Dehaene, S. (2007). Levels of processing during non-conscious perception: A critical review of visual masking. Philosophical Transactions of the Royal Society of London, Series B, Biological Science, 362(1481), 857–875. Kozmová, M., & Wolman, R. N. (2006). Self-awareness in dreaming. Dreaming, 116(3), 196–214. Kriegel, U. (2007). A cross-order integration hypothesis for the neural correlate of consciousness. Consciousness and Cognition, 16, 897–912. Lamme, V. A. F. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neuroscience, 23(11), 571–579. Lamme, V. A. F. (2001). Blindsight: The role of feedforward and feedback corticocortical connections. Acta Psychologica, 107, 209–228. Lamme, V. A. F. (2003). Why visual attention and awareness are different. Trends in Cognitive Sciences, 7(1). Lamme, V. A. F. (2004). Separate neural definitions of visual consciousness and visual attention: A case for phenomenal awareness. Neural Networks, 17, 861–872. Lamme, V. A. F. (2006a). Towards a true neural stance on consciousness. Trends in Cognitive Sciences, 10(11), 494–501. Lamme, V. A. F. (2006b). Zap! Magnetic tricks on conscious and unconscious vision. Trends in Cognitive Sciences, 10(5), 193–195.
C.C. Augustenborg / Consciousness and Cognition 19 (2010) 547–579
579
Lamme, V. A. F., Supér, H., & Spekreijse, H. (1998). Feedforward, horizontal, and feedback processing in the visual cortex. Current Opinions in Neurobiology, 8, 529–535. Lamy, D., Mudrik, L., & Deouell, L. Y. (2008). Unconscious auditory information can prime visual word processing: A process-dissociation procedure study. Consciousness and Cognition, 17, 688–698. Laureys, S., Perrin, F., & Brédart, S. (2007). Self-consciousness in non-communicative patients. Consciousness and Cognition, 16(3), 722–741 [discussion 742– 745]. Lieberman, M. D., Gaunt, R., Gilbert, D. T., & Trope, Y. (2002). Reflection and reflexion: A social cognitive neuroscience approach to attributional inference. Advances in Experimental Social Psychology, 34, 199–249. Llinás, R. R., & Paré, D. (1991). Of dreaming and wakefulness. Neuroscience, 44(3), 521–535. Llinas, R. R., Ribary, U., Joliot, M., & Wang, X. J. (1994). Content and context in temporal thalamocortical binding. In G. Buzaki, R. Llinas, W. Singer, A. Berthoz, & Y. Christen (Eds.), Temporal coding in the brain (pp. 251–271). Berlin, Germany: Springer-Verlag. Lutz, A. (2007). Neurophenomenology and the study of self-consciousness. Consciousness and Cognition, 16(3), 765–767. Martinez, A., Anllo-Vento, L., Sereno, M. I., Frank, L. R., Buxton, R. B., Dubowitz, D. J., et al (1999). Involvement of striate and extrastriate visual cortical areas in spatial attention. Nature Neuroscience, 2(4), 364–369. Meyer, D. E., & Kieras, D. E. (1997). A computational theory of executive cognitive processes and multiple-task performance: Part 1. Basic mechanisms. Psychological Review, 104(1), 3–65. Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology, 11, 56–60. Näätänen, R. (1992). Attention and brain function. Hillsdale, NJ: Lawrence Earlbaum Associates. Näätänen, R., & Picton, R. (1987). The N1 wave of the human electric and magnetic response to sound: A review and an analysis of the component structure. Psychophysiology, 375–425. Näätänen, R., & Winkler, I. (1999). The concept of auditory stimulus representation in cognitive neuroscience. Psychological Bulletin, 125(6), 826–859. Nagel, T. (1974). What is it like to be a bat. The Philosophical Review, 83(4). Navon, D., & Miller, J. (2002). Queuing or sharing? A critical evaluation of the single-bottleneck notion. Cognitive Psychology, 44(3), 193–251. Nelson, D. L., & Goodmon, L. B. (2004). Strengthening the activation of unconsciously activated memories. Memory and Cognition, 32(5), 804–818. Occhionero, M. (2004). Mental processes and the brain during dreams. Dreams, 14(1), 54–64. Öhman, A., & Soares, J. J. (1993). On e automatic nature of phobic fear: Conditioned electrodermal responses to masked fear relevant stimuli. Journal of Abnormal psychology, 102(1), 121–132. Öhman, A., & Soares, JJ. (1994). ‘‘Unconscious anxiety”: Phobic responses to masked stimuli. Journal of Abnormal Psychology, 103(2), 231–240. Owen, A. M., & Coleman, M. R. (2008). Functional neuroimaging of the vegetative state. Neuroscience, 9, 235–243. Parker, K., Balthazard, C., Regehr, C., & Bowers, K. S. (1990). Intuition in the context of discovery. Cognitive Psychology, 22, 72–100. Parmentier, FB. (2008). Towards a cognitive model of distraction by auditory novelty: The role of involuntary attention capture and semantic processing. Cognition, 109(3), 345–362. Reason, J. (1979). Actions not as planned. In G. Underwood & R. Stevens (Eds.), Aspects of consciousness (pp. 67–89). London: Academic Press. Revonsuo, A., & Newman, J. (1999). Binding and consciousness. Consciousness and Cognition, 8(2), 123–127. Rizzolatti, G., Fadiga, L., Gallese, V., & Fogassi, L. (1996). Premotor cortex and the recognition of motor actions. Brain Research. Cognitive Brain Research, 3(2), 131–141. Rosenthal, D. M. (2005). Consciousness and mind. Oxford University Press. ISBN:139780198236962. Rosenthal, D. M. (2008). Consciousness and its function. Neuropsychologia, 46, 829–840. Sahraie, A., Weiskrantz, L., Barbur, J. L., Simmons, A., Williams, S. C., & Brammer, M. J. (1997). Pattern of neuronal activity associated with conscious and unconscious processing of visual signals. PNAS, 94(17), 9406–9411. Sass, A., & Parnas, J. (2003). Schizophrenia, consciousness and the self. Schizophrenia Bulletin, 29(3), 427–444. Schmitz, T. W., & Johnson, S. C. (2006). Self-appraisal decisions evoke dissociated dorsal–ventral aMPFC networks. Neuroimage, 30(3), 1050–1058. Spence, S. A., Brooks, D. J., Hirsch, S. R., Liddle, P. F., Meehan, J., & Grasby, P. M. (1997). A PET study of voluntary movement in schizophrenic patients experiencing passivity phenomena (delusions of alien control). Brain, 120(Pt. 11), 1997–2011. Steyn-Ross, M. L., Steyn-Ross, D. A., & Sleigh, J. W. (2001). Modelling general anaesthesia as a first-order phase transition in the cortex. Progress in Biophysics and Molecular Biology, 85(2–3), 369–385. Supèr, H., van der Togt, C., Spekreijse, H., & Lamme, V. A. (2003). Internal state of monkey primary visual cortex (V1) predicts figure–ground perception. Journal of Neuroscience, 23(8), 3407–3414. Thagard, P., & Aubie, B. (2008). Emotional consciousness: A neural model of how cognitive appraisal and somatic perception interact to produce qualitative experience. Consciousness and Cognition, 17, 811–834. Thompson, E., & Zahavi, D. (2007). Philosophical issues: Phenomenology. In P. D. Zelazo, M. Moscovitch, & E. Thompson (Eds.). The Cambridge handbook of consciousness. Cambridge University Press. ISBN:978052167412-6. Tononi, G., & Edelman, G. M. (1998). Consciousness and the integration of information in the brain. Advances in neurology, 77, 245–280. Tononi, G., & Edelman, G. M. (2000). Schizophrenia and the mechanisms of conscious integration. Brain Research Brain Research Reviews, 31(2–3), 391–400. Tsakiris, M., Hesse, M. D., Boy, C., Haggard, P., & Fink, G. R. (2007). Neural signatures of body ownership: A sensory network for bodily self-consciousness. Cerebral Cortex, 17, 2235–2244. Vahle-Hinz, C., Detsch, O., Siemers, M., & Kochs, E. (2007). Contributions of GABAergic and glutamatergic mechanisms to isoflurane-induced suppression of thalamic somatosensory information transfer. Experimental Brain Research, 176(1), 159–172. Vaitl, D., Gruzelier, J., Jamieson, G. A., Lehmann, D., Ott, U., Sammer, G., et al (2005). Psychobiology of altered states of consciousness. Psychological Bulletin, 131(1), 98–127. van Gaal, S., Ridderinkhof, K. R., Fahrenfort, J. J., Scholte, H. S., & Lamme, V. A. (2008). Frontal cortex mediates unconsciously triggered inhibitory control. The Journal of Neuroscience, 28(32), 8053–8062. Verbruggen, F., & Logan, G. D. (2008). Long-term after effects of response inhibition: Memory retrieval, task goals, and cognitive control. Journal of Experimental Psychology. Human Perception and Performance, 34(5), 1229–1235. Vogeley, K., May, M., Ritzl, A., Falkai, P., Zilles, K., & Fink, G. R. (2004). Neural correlates of first-person perspective as one constituent of human selfconsciousness. Journal of Cognitive Neuroscience, 16(5), 817–827. Volz, K. G., & von Cramon, D. Y. (2006). What neuroscience can tell about intuitive processes in the context of perceptual discovery. Journal of Cognitive Neuroscience, 18(12), 2077–2087. Watanabe, T., Náñez, J. E., & Sasaki, Y. (2001). Perceptual learning without perception. Nature, 413(6858), 844–848. Wegner, D. M. (2002). The illusion of conscious will. Cambridge, MA: MIT Press. ISBN:0262731622. Weiskrantz, L. (1996). Blindsight revisited. Current Opinions in Neurobiology, 6(2), 215–220. Yago, E., Corral, M. J., & Escera, C. (2001). Activation of brain mechanisms of attention switching as a function of auditory frequency change. Neuroreport, 12(18), 4093–4097. Zahavi, D. (2005). Subjectivity and selfhood: Investigating the first-person perspective. The MIT Press. ISBN:0262240505. Zeki, S., & Bartels, A. (1999). Toward a theory of visual consciousness. Consciousness and Cognition, 8(2), 225–259. Zhang, J., Riehle, A., Requin, J., & Kornblum, S. (1997). Dynamics of single neurons activity in monkeys primary motor cortex related to sensorimotor transformation. Journal of Neuroscience, 17, 2227–2246.
Consciousness and Cognition 19 (2010) 580–596
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
A global workspace model for phenomenal and access consciousness Antonino Raffone a,c,*, Martina Pantani b a b c
Department of Psychology, ‘‘Sapienza” University of Rome, Italy Department of Philosophy and Social Sciences, University of Siena, Italy Laboratory for Perceptual Dynamics, Brain Science Institute RIKEN, Japan
a r t i c l e
i n f o
Article history: Received 5 June 2009 Available online 9 April 2010 Keywords: Consciousness Global workspace Attention Model Neural network Phenomenal consciousness Conscious access Feedback
a b s t r a c t Both the global workspace theory and Block’s distinction between phenomenal and access consciousness, are central in the current debates about consciousness and the neural correlates of consciousness. In this article, a unifying global workspace model for phenomenal and access consciousness is proposed. In the model, recurrent neural interactions take place in distinct yet interacting access and phenomenal brain loops. The effectiveness of feedback signaling onto sensory cortical maps is emphasized for the neural correlates of phenomenal consciousness. Two forms of top-down attention, attention for perception and attention for access, play differential roles for phenomenal and access consciousness. The model is implemented in a neural network form, with the simulation of single and multiple visual object processing, and of the attentional blink. Ó 2010 Elsevier Inc. All rights reserved.
1. The global workspace model One of the most currently influential theories of human consciousness, with fundamental implications for addressing the Neural Correlates of Consciousness (NCC), is given by Bernard Baars’ Global Workspace (GW) theory (Baars, 1983, 1998; Baars, Ramsoy, & Laureys, 2003). In this theory conscious perception enables access to widespread brain sources, in terms of broadcasting, whereas in unconscious sensory processing these brain sources process information in a substantially segregated or modular fashion. According to Baars (1998), consciousness, although limited in capacity at any given time, enables an ongoing gateway to extensive unconscious knowledge sources in the brain, therefore creating the conditions for a global access in cerebral information processing. There is wide agreement and remarkable empirical support about a GW dynamics in the brain, related to conscious experience. In particular, Baars’ GW theory has been revisited in a neuronal GW framework by Stanislas Dehaene and collaborators, based on a coherent set of psychophysical, neuroimaging and computational investigations (Dehaene, Kerszberg, & Changeux, 1998; Dehaene, Changeux, Naccache, Sackur, & Sergent, 2006; Dehaene & Naccache, 2001; Gaillard et al., 2009). With reference to visual awareness, this neuronal GW model proposes that unconscious visual information processing is characterized by the parallel activation of multiple modular brain networks (Dehaene et al., 1998; Dehaene & Naccache, 2001; Dehaene, Sergent, & Changeux, 2003; Dehaene et al., 2006; Gaillard et al., 2009). Three conditions have to be met to enable the access to consciousness of incoming visual information (Dehaene et al., 2006; Gaillard et al., 2009): Condition 1: information must be explicitly represented by neuronal firing in perceptual networks in visual cortical areas coding for the specific features of the conscious percept. Condition 2: this sensory neuronal representation must reach a minimal threshold of duration and intensity necessary for access to a second stage of processing involving a distributed cortical network, with * Corresponding author at: Department of Psychology, ‘‘Sapienza” University of Rome, Via dei Marsi, 78, 00185 Rome, Italy. Fax: +39 064462449. E-mail address:
[email protected] (A. Raffone). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.013
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
581
special reference to parietal and prefrontal cortices. Condition 3: through joint bottom-up propagation and top-down attentional amplification, a coherent reverberant state must be ignited in terms of a brain-scale neural assembly, thus implementing a global workspace. These conditions will be revisited in light of the model proposed in next sections of the article. Strength and stability of neural activations at a brain-scale are widely regarded fundamental for conscious access in GW neurodynamics. As suggested by Dehaene et al. (2003), Sergent and Dehaene (2004), these higher strength and stability of neural firing for conscious access would occur with the ignition of reverberating states involving prefrontal cortex and widely distributed neuronal populations in the brain. The neuronal GW model is characterized by a winner-take-all dynamics at higher stages of neural processing (involving prefrontal cortex), a sort of ‘neural bottleneck’, such that only one large-scale reverberating neural assembly is active in the neuronal GW at a given moment. Such winner-take-all processes have been highlighted in experimental and related computational settings using the attentional blink (AB) paradigm, in which the subjects have to report the identity of two target stimuli (e.g. digits) in a series of rapidly presented visual stimuli, most of which are distracters (e.g. letters). If the second (T2) of the two target stimuli is presented within 500 ms of the first one (T1) in a rapid sequence of distracters, it is often not detected, thus resulting in an AB (Dehaene et al., 2003; Raymond, Shapiro, & Arnell, 1992). In light of their experimental evidence and related modeling results, Sergent and Dehaene (2004) observe that the neuronal global workspace model (Dehaene et al., 1998, 2003) provides a neural account of the two stages described in psychological models of the AB (Chun & Potter, 1995). The first processing stage would correspond to the ‘‘feedforward sweep” in visual cortical processing (Lamme, 2003, 2004; Lamme & Roelfsema, 2000), where a stimulus is automatically processed by the sequential bottom-up activation of brain areas. The second stage would correspond to a top-down amplification, based on endogenous attention and long-range recurrent signaling involving feedback connections (Lamme, 2003; Lamme & Roelfsema, 2000). As also demonstrated by Dehaene et al. (2003) simulations, on conscious trials bottom-up and top-down attentional inputs interact to ignite a broad network of brain areas via long-range recurrent connections. However, if the first activation does not reach the dynamic threshold for self-amplification, neural activation is not broadcasted, being confined to a bottom-up transient signal, in the absence of conscious perception of the stimulus. In this all-or-none interpretation, the AB acts by ‘cutting’ the top-down dynamic support for T2 as workspace neurons are temporarily occupied by coding T1 based on their experimental and computational modeling results, Sergent and Dehaene (2004) see also Dehaene et al. (2006) argued that access to consciousness is all-or-none. However, Overgaard, Rote, Mouridsen, and Ramsøy (2006) proposed an alternative view emphasizing the graded character of conscious experiences. In contrast with Dehaene and collaborators’ GW view, however, Lamme (2003, 2004) suggested that top-down attention operates after conscious perception, to enable report. In this article, based on a simulated neurocomputational model, a reconciliation of these two views is proposed, by distinguishing two levels at which top-down attention can operate, with reference to Block’s distinction between phenomenal consciousness and access consciousness, which is considered in the next section. Simulation evidence reconciling the all-or-none (Sergent & Dehaene) view and the graded (Overgaard et al., 2006) view on perceptual awareness, will also be presented.
2. Phenomenal and access consciousness The GW framework has been used by Block to characterize the distinction between phenomenal consciousness and access consciousness (Block, 1995, 2005, 2007). According to Block, phenomenally conscious content is what differs between experiences as of red and green, whereas access conscious content is content information about which is ‘broadcast’ in the GW. Specifically, Block characterizes access conscious contents in terms of information about which is made available to the brain’s ‘consumer’ systems: systems of memory, perceptual categorization, reasoning, planning, evaluation of alternatives, decision making, voluntary direction of attention, and more generally, rational control of action. Since when we view a complex visual scene we experience a richness of content that seems to go beyond what we can report, Block proposes a state of phenomenal consciousness distinct from global access or GW broadcasting. Block’s proposal is also based on the report of participants in experiments with Sperling’s iconic memory paradigm, claiming to see the whole array of flashed letters, although they could later report only one subsequently cued row or column. Along these lines, it has been suggested that access consciousness is related to (relatively) stable working memory representations, and phenomenal consciousness to a transient iconic memory (Block, 2007; Lamme, 2003). However, this proposal has been criticized by claiming that Block’s phenomenal consciousness would just correspond to a preconscious state, and the perceptual awareness experience of viewers in iconic memory experiments to an illusion (Dehaene et al., 2006). In a recent article, Block (2007) has revisited his theory by contrasting reportability with cognitive accessibility: ‘‘I will be talking about cognitive accessibility instead of reportability. Reportability is a behavioristic ladder that we can throw away. In previous papers (Block, 1995, 2005), I have argued that there can be phenomenally conscious states that are not cognitively accessible (I put it in terms of phenomenal consciousness without access consciousness.) But I am mainly arguing for something weaker here. Cognitive accessibility could be a causally necessary condition of phenomenal consciousness without being a constitutive part of it. Bananas constitutively include CH2O molecules but not air and light. Still, without air and light, there could be no bananas – they are causally necessary. The focus here is on whether accessibility is constitutively necessary to phenomenal consciousness, not whether it is causally necessary” (Block, 2007, p. 484). Block (2007), with reference to the GW model, also argues: ‘‘If we suppose that the neural basis of the phenomenology does not include
582
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
the workspace activations, we can appreciate a neural mechanism by which phenomenology can overflow cognitive accessibility” (Block, 2007, p. 497). To characterize cognitive accessibility Block (2007) refers to Dehaene and Naccache (2001; see also Dehaene et al., 2006), who earlier suggested that some information encoded in the nervous system is permanently inaccessible at a conscious level (set I1), other information is in contact with the workspace and could be consciously amplified if it was attended to (set I2), with only a subset of the latter being mobilized into the workspace (set I3). According to Block (2007): ‘‘What it is for representations to be in the workspace (I3) involves both actuality (sent to the workspace) and potential (can be accessed by consuming mechanisms without further processing). The representations that are actually in the workspace are in active contact with the consuming systems, and the consuming systems can (potentially do) make use of those representations. We might speak of the representations in I3 (in the workspace) as cognitively accessible in the narrow sense (in which consuming mechanisms make use of what is already there), and representations in the union of I3 and I2 as cognitively accessible in the broad sense. It is narrow cognitive accessibility that Dehaene et al. identify with phenomenology. When I speak of phenomenology overflowing cognitive accessibility, I mean that the capacity of phenomenology is greater than that of the workspace – so it is narrow accessibility that is at issue” (Block, 2007, p. 492). With special reference to conscious visual perception, Block (2007) emphasizes the crucial role of long-range recurrent interactions between anterior (frontal) and posterior cortical (e.g. parietal) neural assemblies (coalitions) for perceptual experience: ‘‘Dominant coalitions in the back of the head trigger coalitions in the front of the head that themselves compete for dominance, the result being linked front and back winning coalitions” (Block, 2007, p. 497). And to characterize the relationship between GW and phenomenal consciousness, Block (2007) also remarks: ‘‘If we make the former assumption – that workspace activation is not part of the neural basis of phenomenology – we have a mesh between the psychological result that phenomenology overflows cognitive accessibility and the neurological result that perceptual representations that do not benefit from attention can nonetheless be almost as strong (and probably recurrent) as perceptual representations that do benefit from attention”(Block, 2007, p. 498). And finally: ‘‘my conclusion that the neural machinery of cognitive access is not partially constitutive of phenomenology leaves room for causal influence in both directions. And it may be that top-down causal influence is almost always involved in making the phenomenal activations strong enough” (Block, 2007, p. 498). Thus, according to Block, the neural machinery of phenomenal consciousness has a greater capacity than the GW-related cognitive accessibility, although both the GW and the neural machinery for phenomenal consciousness are plausibly based on longrange recurrent interactions between anterior and posterior neural assemblies. 3. Toward a global workspace model for phenomenal and access consciousness To develop our model, Block’s distinction between access and phenomenal consciousness, and his recent revisitation in terms of cognitive accessibility and long-distance recurrent mechanisms for both narrow and broad cognitive accessibility, are endorsed. We also endorse Block’s view about the possibility of mutual causal influences between a GW-based access consciousness and a phenomenal consciousness with a greater capacity. It is plausible that narrow sense cognitive accessibility demands a high involvement of prefrontal cortex (see also Dehaene & Changeux, 2005; Dehaene et al., 2003), in terms of sustained recurrent signaling with executive ‘consumer system’ neurons. This narrow sense cognitive accessibility is linked to a serial GW dynamics of conscious access as suggested by the AB phenomenon and the PRP (Psychological Refractory Period) effect (Edelman & Tononi, 2000; Pashler & Johnston, 1998). Broad sense cognitive accessibility would imply a recurrent neural signaling in parallel cortico-cortical and thalamo-cortical loops (e.g. in the visual system), with a more limited involvement of prefrontal cortex. In the visual modality, this recurrent signaling might be linked to a transient visual short-term memory store with a larger capacity than the visual working memory store (Lamme, 2003; Landman, Spekreijse, & Lamme, 2003; Sligte, Scholte, & Lamme, 2008). In the following sections the functional logic behind our model will be outlined, and specified in a recurrent multi-layer neural network form, with related simulations, in the next sections of this article. 3.1. The effectiveness of cortical feedback to sensory maps In our model, long-range recurrent neural signaling characterizes both access and phenomenal consciousness. Feedback signaling arriving to sensory (e.g. visual) maps originates from both neural elements linked to the winner-take-all GW assembly (e.g. Dehaene et al., 2003) and ‘loosing’ coalitions (e.g. neural assemblies in parietal posterior cortex in competition with the assembly which is currently accessed in the GW). Indeed, all these assemblies would be capable to back-project to neuronal groups in sensory (e.g. visual) cortical areas coding for the specific features of percepts (e.g. V1). We underline that this cortical feedback signaling is characterized by phenomenal neural effectiveness, as it changes neuronal firing in sensory maps assumed to represent explicitly the features of conscious percepts (Gaillard et al., 2009). Thus, remarkably, feedback signals from cortical assemblies linked to both narrow (‘winning coalition’) and broad cognitive accessibility (‘loosing coalitions’) are potentially effective on a key neural substrate of the phenomenology of perceptual experiences, in line with Block’s (2007) claim that phenomenology overflows (narrow) cognitive accessibility. This view appears to reconcile Lamme and Roelfsema’s (2000) suggestion that virtually any recurrent loop, even locally within visual cortex, might suffice to cause conscious-level processing, with the evidence about a ‘late’ conscious access as recently revealed by intracranial markers (Gaillard et al., 2009).
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
583
We suggest that the winner-take-all GW dynamics for access consciousness involves a GW gate, in terms of an assembly of frontoparietal neurons acting as a gateway to link consumer systems with perceptual representations at different levels for report, decision, evaluation and reasoning. We refer to this aspect as to access neural effectiveness, which is neuro-operationally expressed by the bias in distributed firing in a global neural assembly for conscious access, due to recurrent signal exchanges with executive consumer system neurons, via the GW gate assembly. The GW gate would thus link the neural loop machineries for narrow and broad cognitive accessibility. The present notion of GW gate assembly corresponds to the higher-level neuronal groups with competitive interactions simulated by Dehaene et al. (2003). 3.2. Two forms of top-down attention for consciousness In our present model, we distinguish two complementary forms of top-down attention for consciousness: attention for perception and attention for access. Such a distinction relates to an earlier distinction between filtering (here related to attention for perception) and pigeonholing (here related to attention for access), in a formal Theory of Visual Attention (TVA; Bundesen, 1990), and in a later development of TVA with a Neural Theory of Visual Attention (NTVA; Bundesen, Habekost, & Kyllingsbaek, 2005). These two complementary forms of top-down attention can be clarified with reference to partial report studies, in which subjects are asked to report a subset of visual objects appearing in a briefly presented visual display, based on a selection criterion (e.g. Duncan, 1984; Sperling, 1960). For example, in a partial report task, in which the category of black letters in a briefly presented visual display with black and white alphanumeric characters, given by letters and digits, has to be reported, attention for perception would correspond to the perceptual processing enhancement of black visual objects, and attention for access to the bias for report (access) of the letter rather than digit categories. In another example, in a task in which only visual objects briefly presented on the left or the right of a fixation point are relevant for report or change detection, whereas visual objects on the other side of the fixation point are not, attention for perception would correspond to the perceptual processing enhancement of the objects whose locations are on the cued side, and attention for access to the bias for report (access) of the relevant categories. In another further example, in a visual working memory task based on change detection (Vogel, McCollough, & Machizawa, 2005), in which orientation is the relevant dimension for the change or no-change response, but only red bars in a mixed display of red and blue bars are relevant for the detection report, attention for perception would operate on colors, and attention for access on orientations. It is also possible to conceive a mirror task of the latter, in which attention for perception operates on orientations (e.g. only vertical bars are relevant for the detection report), and attention for access on colors (i.e., color is the relevant dimension to be accessed for change detection), with displays of colored bars with vertical and horizontal orientations. In our model, top-down attention for perception operates primarily at an ‘‘intermediate level”, plausibly corresponding to parietal posterior cortex (for spatial coding) or lateral occipital complex (for object feature coding) and associative areas such as V4 in the visual modality (e.g. Behrmann, Geng, & Shomstein, 2004; Murray & Wojciulik, 2004; Sligte et al., 2008). It affects phenomenal consciousness by back-projections to sensory maps and arising recurrent interactions (phenomenal loop), and access consciousness by forward connections to prefrontal cortex, and arising recurrent interactions (access loop). Topdown attention for access is directly implied in access consciousness in prefrontal cortex. Also top-down attention for perception can serve access consciousness. However, whereas access consciousness demands attention for access and thus neuronal firing in prefrontal cortex (ignition of consumer systems) to be established, attention for perception operates by enhancing perceptual representations, also to serve access consciousness when necessary. Moreover, attention for perception may not be necessary for access consciousness, for example when a conscious access for report is based on verbal working memory (Baddeley & Hitch, 1974).
4. Modeling global workspace processes for phenomenal and access consciousness Given the model framework of GW interactions for phenomenal and access consciousness presented in the previous section, here we present a computational model to simulate neural correlates of narrow and broad cognitive accessibility (Block, 2007) and related attentional processes, in a parallel distributed GW neurodynamics. This model aims to show in a transparent fashion plausible neurocomputational bases for phenomenal and access consciousness, rather than to capture detailed neuroanatomical and neurophysiological properties as in other models (e.g. Dehaene et al., 2003; Lumer, Edelman, & Tononi, 1997a, 1997b). The more technical aspects of the computational model, enabling a replication of the performed simulations, are presented in the Appendix A. 4.1. Neural network architecture and units The neural network architecture includes four recurrently-connected layers to simulate the following functional brain components, from bottom (see Fig. 1): (1) a sensory map (e.g. V1–V4 in the visual domain), where sensory inputs are explicitly represented;
584
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
Fig. 1. Neural network architecture. The phenomenal loop is established between the sensory map and the phenomenal register layers. The access loop is established between the GW gate and the consumer systems layers. The link between phenomenal loop and access loop is established between the phenomenal register and the GW gate. Note the forward and feedback connection between pairs of layers, the top-down attentional inputs to phenomenal register (attention for perception) and consumer system (attention for access) layers from implicitly modeled prefrontal areas. The sensory map receives sensory input. See text for more explanations.
(2) a phenomenal register, which for the visual modality might plausibly correspond to parietal posterior cortex and visual associative areas in the temporal and occipital lobes (such as area V4), at an intermediate level between sensory maps and prefrontal cortex (GW gate), assumed to represent ‘loosing’ neuronal coalitions for broad cognitive accessibility (Block, 2007); (3) the GW gate, corresponding to a winner-take-all processing ‘bottleneck’, plausibly associated to executive control areas such as anterior cingulate cortex, dorsolateral prefrontal cortex and superior parietal cortex (Dehaene et al., 1998, 2003; Owen, Evans, & Petrides, 1996; Wager & Smith, 2003); (4) the consumer systems, involved in conscious access broadcasting (Block, 2005, 2007), with reference to a mosaic of prefrontal regions involved in decision making, evaluating, planning and reasoning (e.g. Miller & Cohen, 2001). Note that such prefrontal areas can interact with other cortical and subcortical regions in conscious access (broadcasting). In order to reflect in a simplified fashion the phenomenology of both neural activation levels (related to neuronal firing rates) and spike timing (neuronal oscillations), considered to be both relevant in visual attention (e.g. Buehlmann & Deco, 2008; Fries, Reynolds, Rorie, & Desimone, 2001), visual working memory (Raffone & Wolters, 2001), visual awareness (e.g. Dehaene et al., 2003; Edelman & Tononi, 2000; Gaillard et al., 2009), two kinds of excitatory neural units are implemented in the neural network (Fig. 2): slow and fast excitatory neural units. The slow units are ‘leaky-integrators’, often used in the connectionist literature (e.g. McClelland & Rumelhart, 1986); each fast excitatory neural unit, coupled to an inhibitory unit, behave as a ‘relaxation oscillator’, used to model visual psychophysical and neurophysiological findings (Grossberg & Grunewald, 1997; Grossberg & Somers, 1991) (see the Appendix A for the related equations). All such neural units correspond to local groups of neurons. The fast excitatory neural units might stand for groups of neurons with prevalently accommodative response properties, i.e. neuronal groups composed of morphologically simple pyramidal cells with spike train accommodation, for transient cortical responses (see Wang et al., 2006). The slow excitatory neural units might stand for groups of neurons composed of complex pyramidal cells displaying non-accommodating discharge behavior, for sustained cortical responses (see Wang et al., 2006). In terms of synaptic current characteristics, the fast excitatory neural units might reflect ‘fast’ AMPA-mediated local synaptic current exchanges, whereas the slow excitatory neural units might reflect ‘slow’ NMDA-mediated; the inhibitory unit coupled with the fast excitatory unit might stand for a group of inhibitory cells with a GABAa-mediated synaptic output (see Wang, 1999, about a neural network model for neural bases of working memory with an explicit modeling of such synaptic currents). The excitatory nodes of the fast neural units in the GW gate layer receive an inhibitory signal from a linked node with a slow rise of activation, to model slow neuronal firing adaptation. A similar adaptation mechanism was used by Dehaene et al. (2003) to simulate the decay of GW activity for a given target object (T1 in the simulated AB task).
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
585
Fig. 2. Diagram of connectivity between neural units within and between network layers. Lines with arrow ending stands for voltage-independent excitatory connections, lines with diamond endings for voltage-dependent excitatory connections, and lines with circle endings for inhibitory connections. Oscillations arise by the coupling of fast excitatory units with a paired inhibitory unit. Voltage-independent excitatory connections between fast excitatory units coding for the same object in a given layer (with a synchronizing effect), and inhibition for competition via inhibitory units activated by fast excitatory units coding for different objects in a given layer, are not displayed. See text for more explanations, and the Appendix for the computational schemes.
The sensory map layer, which is mainly characterized by transient neural responses only includes clusters of fast excitatory and coupled inhibitory units. The other three layers include clusters with three neural units of the different kinds (see Fig. 2). As in Dehaene et al. (2003), we assume that each visual object is encoded by a stream of units across the four network layers. In the present model, each object and related task setting (e.g. consumer system codes) is thus encoded by 10 neural units of each kind in the layers. Except for the sensory map, the excitatory node of each fast neural unit is coupled to a slow (excitatory) neural unit. Thus, 110 neural units (40 fast excitatory, 40 inhibitory and 30 slow excitatory) are allocated to each simulated visual object (object stream) in the neural network. For the sake of simplicity, as in Dehaene et al. (2003), there is no overlap between object streams. For the intra-layer connectivity (see Fig. 2), the fast excitatory units coding for the same object are mutually connected to provide a intra-area synchronization effect (see also Grossberg & Grunewald, 1997). In order to provide a local neural circuit for reverberating activity, except for the sensory map, the fast excitatory units are reciprocally connected with a paired slow excitatory unit. To model competition between neural object representations, except for the sensory map (where neuronal receptive fields are small), there is another inhibitory unit ‘for competition’ activated from a linked fast excitatory unit, and sending inhibition to all fast excitatory nodes for other objects in the same layer. In the GW gate layer, this inhibition for competition allows only the neural units of one object to be active at a given time. For the inter-layer connectivity (see Fig. 2), there are forward connections from the sensory map to the phenomenal register, and from this to the GW gate, which in turn sends forward connections to the consumer system layer; feedback connections are established from top to bottom between layer pairs, mirroring the forward connectivity between layers. For the sake of simplicity, there is a one-to-one connectivity scheme between neural units in different layers in forward and feedback coupling, which results in 10 forward and 10 backward connections from/to each layer, for each object. The forward connections are established between fast excitatory units, whereas the backward connections are established from slow to fast excitatory units. Feedback connections are plausibly voltage-dependent, i.e. their effect depends on the activation level of the receiving units, thus acting in a modulatory fashion (Dehaene et al., 2003; Lumer et al., 1997a, 1997b; Tononi, Sporns, & Edelman, 1992). See the Appendix A for the related computational scheme. Attention for perception is implemented by a voltage-dependent input to fast excitatory neural units in the phenomenal register, from an implicitly modeled prefrontal area for goal-related (task-related) encoding of perceptual relevance. Attention for access is implemented by a voltage-dependent input to fast excitatory neural units in the consumer system layer, from an implicitly modeled prefrontal area for goal-related (task-related) encoding of access relevance. In prefrontal cortex this voltage-dependent modulation might involve NMDA currents, which can be affected by dopamine (Brunel & Wang, 2001). A similar implicit modeling of the neuronal population sources in prefrontal cortex for top-down attentional signals, has been conducted in several biologically-inspired computational modeling studies (e.g. Corchs & Deco, 2004; Usher & Niebur, 1996).
586
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
4.2. Simulations First of all, we studied the multi-layer neural network response to a single object presentation, in different attentional conditions. Stimuli were presented for the entire duration of the simulations, corresponding to 400 ms of real time. The average activity levels of fast and slow excitatory neural units in each layer for each object were measured and displayed. Note that although for immediacy of demonstration and report space limitations, in this report we show single simulations for each visual presentation condition, our results appear appreciably robust to changes in parameters and random generation seeds (i.e. initial values and noise time-series). However, a more systematic exploration of neural network dynamics is beyond the aims of this article. As shown in Fig. 3, with both attention for perception and attention for access, high-frequency oscillatory activity (in the so-called gamma band; see Lumer et al., 1997a, 1997b, and Tononi et al., 1992) is observed at all layers in terms of fast excitatory activity levels. It can be clearly seen that at each layer the fast excitatory units coding for the same object are synchronized in their oscillations arising from their coupling with paired inhibitory units, via the synchronizing connections. Note that the flow of activity waves proceeds from the bottom sensory map layer at each oscillatory cycle. Correspondingly high average activity levels for slow excitatory units are observed in the three upper layers. Note the slower ignition of high-frequency (synchronous) oscillatory activity at the top consumer system layer. We then replicated the same simulation as in Fig. 3, but without attention for perception, i.e. setting equal to 0 the corresponding voltage-dependent input to neural units at the phenomenal register layer. This second simulation is displayed in
Fig. 3. Neural network activity with presentation of a visual object, with both attention for perception and attention for access. In this figure and in the following ones, in each layer, the black lines display the average activity of the 10 fast excitatory units coding for the object in the layer; the gray lines display the average activity of the 10 slow excitatory units coding for the object in the same layers (except for the sensory map layer). See text for more descriptions and explanations.
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
587
Fig. 4. Note the dramatic decreases of neural activity levels at the sensory map and especially at the phenomenal register, and the absence of any activity at the GW gate and consumer system levels. This simulation evidence can be closely related to Dehaene et al. (2006, p. 206) ‘‘2 2” array about the necessity of joint sufficiently-high stimulus strength and top-down attention for spreading of neural activity to anterior brain areas, and conscious access. Then, we replicated the same simulation as in Fig. 3, but this time without attention for access, i.e. setting equal to 0 the corresponding voltage-dependent input to neural units at the consumer systems layer. This third simulation is displayed in Fig. 5. Note that the network exhibits a different neurodynamical behavior as compared to both simulations with both forms of attention (Fig. 3), and without attention for perception (Fig. 2). Specifically, neural activity at the sensory map is minimally affected as compared to the condition with both forms of attention, whereas neural activity in the access loop, i.e. at the GW gate and consumer systems layers, is dramatically reduced or nearly-absent. Note also that in this simulation condition neural activity at the phenomenal register layer is mildly but evidently reduced as compared to the condition with both forms of attention, in terms of fast and slow activity, with a related reduction of oscillatory and synchronous neural activity. A comparison between the simulations with and without attention for access (Figs. 3 and 5), sheds light on the notion of access neural effectiveness proposed above. In Fig. 6, the activity levels at the sensory map in the three simulation conditions described above, are presented for a comparative visual inspection. Note the clear reduction of neural activity in the absence of attention for perception (Panel B), and virtually the same level of activity with both forms of attention (Panel A) and without attention for access (Panel C). A comparison between the simulations with and without attention for perception, with the implications for neural activity levels at sensory maps (see Figs. 3, 5 and 6), sheds light on the notion of phenomenal neural effectiveness proposed above.
Fig. 4. Neural network activity in the absence of attention for perception. Note the absence of activity in the access loop layers, and the low activity at the phenomenal register layer. See text for more descriptions and explanations.
588
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
Fig. 5. Neural network activity in the absence of attention for access. Note the relatively high activity levels in the phenomenal loop layers as compared to the access loop layers, with a low activity level at the GW gate layer, and an almost absent activity at the consumer system layer. See text for more descriptions and explanations.
Fig. 6. Display of the sensory map activity level (average activity level of fast excitatory units), in the three simulation conditions shown in Figs. 4, 5 and 6. (A) With both attention for perception and attention for access; (B) without attention for perception; (C) without attention for access. Note the significant reduction of activity level without attention for perception as compared to both the other conditions. See text for more descriptions and explanations.
We then simulated a condition with two simultaneously presented visual objects (T1 and T2), displayed in Fig. 7. The same network and parameters as in the single object simulations, were used. The simulation lasted 400 ms. Note that unlike at the GW gate, where a winner-take-all dynamics is observed, at the phenomenal register parallel neural activations for both T1 and T2 are observed, though the phenomenal register activity level for T1 is higher as compared to T2, due to
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
589
Fig. 7. Neural network activity with simultaneous presentation of two visual objects relevant for access (targets) T1 and T2, both with attention for perception and attention for access. Note the winner-take-all dynamics at the GW gate, recurrent with the consumer system layer in the access loop. Note the parallel neural activities for both objects in the phenomenal loop, in phenomenal register and sensory map layers. See text for more descriptions and explanations.
feedback from the GW gate. The ‘symmetry-breaking’ for the winner-take-all activation at the GW gate takes place due to noise. This simulation appears related to Block’s (2007) view about the coexistence of the GW winner and loosing (neural) ‘coalitions’ in posterior cortical areas (e.g. parietal posterior cortex), recurrently connected with more anterior (prefrontal) cortical areas (see also Section 2). Finally, we simulated a basic attentional blink (AB) effect. The AB simulations last 900 ms. As in Dehaene et al. (2003), for the sake of simplicity, the neural network is presented only with the first target (T1) and the second target (T2), in serial presentation. To model an AB (trial) condition, T2 is presented 300 ms after the onset of T1 (t = 0). This simulation condition is compared with a simulation condition in which T2 is presented 600 ms after the onset of T1. Note that for these AB simulations the same network for the single object presentations (Figs. 3–6) and parallel object presentation (Fig. 7), is used (with the same parameters). Sensory inputs for T1 and T2 to the sensory map layer last for 100 ms. Fig. 8 shows three forms of network layer response with T2 presented 300 ms after T1, as related to an AB trial. At the sensory map, only a transient stimulus-related response is observed for both T1 and T2. At both the phenomenal register and GW gate layers, T1-related neural activity is sustained for an ‘‘AB time-window” of about 500 ms after offset of T1. As in Dehaene et al.’s (2003), this reverberating activity is supported by intra-layer (area) and inter-layer (area) recurrent interactions, and decays due to neural activity adaptation. Note that in this simulation condition, with reference to T1-related responses, a layer in the access loop (GW gate) and a layer in the phenomenal loop (phenomenal register), work in concert. Due to the winner-take-all competitive interactions at the GW gate, T2-related neural activity in the access loop (GW gate and consumer systems layers) is virtually absent. To dissociate access and phenomenal loops, with reference to T2-related responses, a transient neural activation during T2 presentation is however observed at the phenomenal register. Finally, note that the consumer systems layer, is capable to maintain neural activity for T1 after the decay of T1-related activity in the winner-take-all GW gate, thus enabling access to or report of T1 afterwards. Fig. 9 shows the simulation of a trial in the absence of AB, with T2 presented immediately after the AB time-window (600 ms after T1 onset). Notice that in this simulation condition neural activity for T2 is ignited across the three upper layers
590
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
Fig. 8. Simulation of an attentional blink (AB) trial, with T2 presented 300 ms after onset of T1. Note the stimulus-related activation in the sensory maps for both objects, the short-term reverberation at the GW gate and phenomenal register linked to the AB, and maintenance for report of T1 at the consumer systems layer. See text for more descriptions and explanations.
of the network, as for T1. T1 and T2 are both maintained at the consumer systems layer, and are thus available for access (report). 5. Discussion 5.1. Two interacting loops for consciousness Our GW model for phenomenal and access consciousness, which has been implemented and simulated in a multi-layer recurrent neural network form, includes an access loop and a phenomenal loop, in mutual interaction (Fig. 1). Schematically, the access loop is constituted of a GW gate and the consumer systems in conscious access. The phenomenal loop comprises a phenomenal register and sensory maps. The GW gate provides a serial dynamical linking on a brain-scale between consumer system states and perceptual states arising in the phenomenal loop. The dissociation between GW gate and consumer systems, such as working memory systems, is clearly illustrated in our AB simulations (see Figs. 8 and 9). Whereas the GW gate displays a serial winner-takeall dynamics on T1 and T2, in the consumer system layer (acting as a working memory system), more target neural representations can be simultaneously maintained, and then enable report of both T1 and T2 in the AB task. Note however that the consumer systems in conscious access are not restricted to working memory, though working memory plays a crucial role in higher-level cognitive processing domains, including reasoning, planning and decision making (e.g. Baddeley, 2000; Baddeley & Hitch, 1974). It has to be noted that the GW gate, which in our simulations receives input from the visual modality, and sends outputs to a consumer system linked to such visual inputs, can receive inputs on a global brain-scale (e.g. inputs in other sensory modalities) by a massive convergence, and send outputs on a global brain-scale (to a large set of consumer systems) by a massive divergence. It acts as a winner-take-all processing ‘bottleneck’, and plausibly includes a distributed neuronal population in executive control areas such as anterior cingulate cortex, dorsolateral prefrontal cortex and superior parietal cortex
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
591
Fig. 9. Simulation of a trial without attentional blink (AB), with T2 presented 600 ms after onset of T1. Note the stimulus-related activation in the sensory maps for both objects, the sequential short-term reverberations for T1 and T2 at the GW gate and phenomenal register, and maintenance for report of both T1 and T2 at the consumer systems layer. See text for more descriptions and explanations.
(Dehaene et al., 1998, 2003; Owen et al., 1996; Wager & Smith, 2003), and related cortico-thalamic loops (Edelman & Tononi, 2000). The phenomenal register (Fig. 1), plausibly associated to parietal posterior cortex and connected associative cortical areas, links the serial GW gate, characterized by a winner-take-all dynamics, with multiple sensory maps (e.g. areas V4 and V5 in the extrastriate visual cortex, in turn connected to V3, and then to V2 and V1). As shown by our modeling (see Fig. 6), in the phenomenal loop cortical feedback signaling is characterized by phenomenal neural effectiveness, as it changes neuronal firing in sensory maps assumed to represent explicitly the features of conscious percepts (Gaillard et al., 2009). In the access loop, we refer to a complementary access neural effectiveness, which is neuro-operationally expressed by the bias in distributed firing in a global neural assembly for conscious access, due to recurrent signal exchanges with executive consumer system neurons, via the GW gate assembly (see Fig. 5). The importance of effectiveness to characterize the emergence of integrated neural states in a distributed cortical system was emphasized by Tononi et al. (1992) in a large-scale computational model of the visual system. Specifically, Tononi et al. linked integrated neural states to a simulated behavioral response. In another computational modeling study, Lumer et al. (1997b) found that the perturbation of neural synchrony was effective on neuronal firing rates in a large-scale simulation of a thalamo-cortical network. Our present model leads to the prediction that perturbing neuronal firing activity in ‘loosing’ coalitions for phenomenal consciousness (e.g. in parietal posterior cortex) would affect neuronal firing in sensory maps for the explicit encoding of conscious perceptual experiences (Gaillard et al., 2009). This neural effectiveness via feedback connections would argue against the claim of an ‘inert’ preconscious state moved against the notion of phenomenal consciousness, as in Dehaene et al. (2006). 5.2. Attention for perception and attention for access In the present model, we distinguish between two forms of top-down attention: attention for perception and attention for access. For example, these two forms of attention would operate differentially for ‘‘what we see” and ‘‘what we keep in mind”, or, in other words, for broad sense and narrow sense cognitive accessibility (Block, 2007). Our distinction between
592
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
the two forms of attention might shed light on the roles of top-down attention in access and phenomenal consciousness (Block, 2007; see also Lamme, 2003). Attention for perception is plausibly related to the top-down modulation of neuronal responses in perceptual cortices (e.g. in the visual cortex), affecting the latency and firing rates (or spike synchrony) of neurons responding for selected locations, features and objects (e.g. Bundesen et al., 2005; Desimone & Duncan, 1995; Fries et al., 2001). Attention for perception operates by a dynamic allocation of a limited (e.g. visual) processing capacity (Bundesen et al., 2005). For example, if red objects have current priority, and four of them are simultaneously displayed with four blue objects, their attentional weight (Bundesen, 1990; Bundesen et al., 2005) is higher than the attentional weight of the blue objects, i.e. they are allocated a higher (visual) processing capacity than the blue objects. Note in the example that attention for perception, although selective for color, can be distributed between multiple objects. The higher attentional weight of objects, such as the red objects in the example, however, does not imply that they are (all) access-relevant (i.e. a subset of them may not belong to a category relevant for conscious access and related report, such as digits versus letters). Indeed, attention for access is linked to the topdown setting of consumer systems for conscious access. Thus, attention for perception affects the perceptual dynamics for both the phenomenology of perception and conscious access to the enhanced percepts, in posterior perceptual cortices, whereas attention for access enables access itself, and the related report, in the frontal lobes. In the example above, both the phenomenal (perceptual) experience of the red objects, as related to a feedback-based enhancement at sensory maps (phenomenal neural effectiveness, see above) and their concomitant enhanced availability to access (report), are affected by attention for perception. Note, however, that in principle (top-down) attention for perception is not necessary for phenomenal consciousness (see also Block (2007), for a similar view), whereas attention for access is necessary for access consciousness and related report. Moreover, attention for perception can be needed for access consciousness if enhanced perceptual states (e.g. for a selected visual feature) are demanded for a given report. Although in our simulations we have simplified attention for access by a voltage-dependent input to the consumer system layer, in the brain attention for access might plausibly be mediated by the selective allocation of adaptive coding neurons in lateral prefrontal cortex to access-relevant objects. As suggested by Duncan (2001; see also Duncan & Miller, 2002, and Raffone & Srinivasan, 2009), throughout much of prefrontal cortex (with special reference to the lateral areas) the response properties of single cells are highly adaptable, as any given cell has the potential to be driven by many different kinds of input via a dense network of associative synapses, in a task-sensitive fashion. In a different view, Bundesen et al. (2005) suggested that filtering (related to attention for perception in our model) is mediated by the allocation of neurons to visual objects, and pigeonholing (related to attention for access in our model) to firing rate bias, without involvement of prefrontal cortex. However, further investigations are necessary to shed light on the role of adaptive coding neurons in prefrontal cortex in attention for access as proposed here. According to Block, amplification of neural firing by top-down attention is not constitutively necessary for phenomenal consciousness, ‘‘although top-down causal influence is almost always involved in making the phenomenal activation strong enough” (Block, 2007, p. 498). In light of our model, attention for access is necessary to activate the access loop (Fig. 5), with a crucial role of prefrontal cortex. The neurotransmitter dopamine is likely to play a key role in the access loop, at both levels of the consumer systems (e.g. working memory; see Cohen, Braver, & Brown, 2002) and GW gate (Colzato, Slagter, Spapè, & Hommel, 2008). Attention for perception, on the other hand, can affect phenomenal consciousness by amplifying neural firing in brain regions intermediate between sensory maps and prefrontal cortex, such as parietal posterior cortex (see Fig. 4). As considered above, however, top-down attention for perception also plays a role in conscious access linked to perception. Related to this aspect, Cartwright-Finch and Lavie (2006) found that conscious perception of task-irrelevant stimuli with an inattentional blindness paradigm critically depends upon the level of task-relevant perceptual load. The present model accounts for this evidence in terms of limitations of attention for perception affecting the access-based report of the task-irrelevant stimuli. Attention for perception may be relatively focused or distributed depending on task conditions (Raffone & Srinivasan, 2009; Srinivasan, Srivastava, Lohani, & Baijal, 2009), and the nature of processed information (Treisman, 2006). Finally, it has to be noted that other factors, a part from the ‘enabling’ condition of vigilance (Dehaene et al., 2006), such as the alerting network of attention (Posner & Fan, 2004), can affect both phenomenal and access consciousness, in a relatively unselective mode. Finally, a higher ‘‘cortical tone” or ‘‘vigilance level” for a system of posterior cortical areas (e.g. in extrastriate visual cortex), may be set by ascending subcortical modulation recruited by prefrontal cortex in a given task context, if attention for perception in a given sensory modality is implied (see Luria (1974), for an integrated model of brain function with prefrontal cortex controlling subcortical centers for the regulation of cortical system tone by distributed ascending signals). 5.3. All-or-none versus graded conscious representation Sergent and Dehaene (2004; see also Dehaene et al., 2003) investigated whether the AB degrades the clarity of T2 or whether it corresponds to an ‘all-or-none’ loss of T2 conscious perception. Subjects were asked to rate on a continuous scale how visible T2 was, and then to report T1 identity. The crucial evidence related to perceptual awareness was that the subjects almost never used the intermediate scale values in their visibility judgement of T2, as the ‘blinked’ stimulus was rated as totally unseen or as totally seen in most trials. In a second masking experiment, however, subjects rated visibility in a more continuous manner than in the first AB experiment (Sergent & Dehaene, 2004). Finally, in a third experiment the
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
593
AB and masking procedures were combined. The AB was still found to give rise to all-or-none visibility response patterns. According to Sergent and Dehaene, their experimental findings together suggest that perceptual awareness is characterized by an all-or-none dynamics. Indeed, Dehaene et al. (2003) predicted such an all-or-none GW dynamic behavior by computer simulations. Overgaard et al. (2006) observed, however, that the three experiments run by Sergent and Dehaene lead to two methoddependent results and implications: with the AB procedure subjects tend to respond in an all-or-none fashion, whereas when using a masking procedure subjects respond in a continuous manner. They also noted that whether we should conceive of conscious perception as an all-or-none or a continuous phenomenon, a logical answer is that if subjects report degrees of conscious awareness in some but not all kinds of experimental set-up, this is sufficient evidence that conscious perception is a continuous phenomenon. By using a four-point Perceptual Awareness Scale (PAS), with an emphasis on an introspective report of the perceptual experience rather than on report of stimulus features as in Sergent and Dehaene’s instructions, Overgaard et al. found experimental evidence consistent with different ‘kinds’ or thresholds of conscious perception. Our simulations show the concurrency of both all-or-none states, related to the winner-take-all GW gate activity and the linked non-linear inter-area recurrent activation, and graded neural activations in the phenomenal register (e.g. parietal posterior cortex), sensory maps (e.g. V1–V4) and the consumer systems (e.g. lateral prefrontal cortex regions), depending on the strength of attentional and sensory inputs to such areas (see in particular Figs. 5 and 7 versus Fig. 3). The pronounced nonlinearity and ignition dynamics observed in our model are reminiscent of the more neurophysiologically detailed model of Dehaene et al. (2003; see also Dehaene & Changeux, 2005). This convergence appears remarkable given the different model networks that we used. It has to be noted that we used a lower mutual inhibition level at the phenomenal register layer, in between the sensory maps with no mutual inhibition and the GW gate with a high inhibition level for the winner-take-all competition, and (voltage-dependent) top-down attentional signals to the phenomenal register and the consumer system layers. It remains to be explored whether our simulation findings hold in a more complex and realistic large-scale neural network model.
5.4. Further computational modeling and experimental studies Given the methodological difficulties in experimental testing of phenomenal consciousness (Dehaene et al., 2006), and the likely complexity of the underlying neurodynamics, the testing of hypotheses deriving from the present model might require inferences from different sources of evidence in neuro-cognitive research. The model might provide a unifying framework for such inferences. To test directly the neural processes and mechanisms for phenomenal consciousness as in the present model, electroencephalographic recordings and event-related potentials (ERPs) from multiple brain sites during a Sperling’s iconic memory task, can be correlated to both the self-monitored phenomenal awareness and the explicit report of the briefly presented alphanumerical display. This approach can also be used with presentation of natural visual images and estimates for different levels of visual awareness, as in the study by VanRullen and Koch (2003). In light of our model, it appears interesting to explore electroencephalographic indices (e.g. synchrony-related) of perceptual switching in binocular rivalry and bistable percept experiments (Blake & Logothetis, 2002), in posterior and anterior cortical sites, to characterize differentially and interdependently the neurodynamics of phenomenal and access loops related to perceptual switching phenomenology and access-based report. Specifically, it is predicted that during a transient in the switch from one perceptual interpretation to the other, non-local global assemblies for the two assemblies can coexist in posterior cortical sites (e.g. parietal posterior cortex), with a sharper winner-take-all neurodynamics in prefrontal sites. A direct extension of our present neural network would allow an explicit modeling of the goal-related neuronal populations in prefrontal cortex, in an Attention Setting layer, controlling the patterns of top-down attention signals for access and attention for perception. Such neurons would be for example activated by external instructions (cues) and capable of selfsustained activity over a block of trials. Interestingly, in a recursive dynamics of attentional setting and conscious access, the neural activations in the modeled consumer systems during conscious access, would be capable to change the activation pattern in the Attention Setting layer, recursively affecting top-down attentional signals and thus attention for perception and attention for access. Inspired by our present neural network model, further computational modeling investigations with large-scale corticocortical and thalamo-cortical loops characterized by biological and cognitive constraints (see Dehaene et al., 2003; Lumer et al., 1997a, 1997b; Tononi et al., 1992), would also contribute to shed light on the interacting neural processes and mechanisms for phenomenal and access consciousness proposed here. Simulation of cognitive tasks differentially involving access consciousness, phenomenal consciousness, attention for access and attention for perception, can be implemented, with the observation of large-scale neural assembly dynamics in different task conditions. Such large-scale neural network models can be constrained in terms of realistic key anatomical structures and neurophysiological parameters, and of simulated cognitive performance such as in the attentional blink, iconic and visual working memory tasks. In these neurocomputational investigations a perturbation-based approach, as in the study by Lumer et al. (1997b), can be carried out, with the observation of how the simulated perturbation neural synchrony and neuronal firing rate patterns affect the dynamics of neural assemblies for narrow and broad cognitive accessibility.
594
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
Also the neurophenomenology approach (Lutz & Thompson, 2003; Varela, 1996) can contribute to shed light on the neurocognitive processes for phenomenal and access consciousness that we have highlighted. In the neurophenomenology approach quantitative measures of neural activity are combined with first person data about the subject’s experience. Participants’ reports can thus be useful in identifying variability in brain activity from moment to moment; this unique information might guide the detection and interpretation of neural processes correlated to different aspects of conscious experience. As the mental practice of meditation involves techniques with a differential emphasis on access consciousness, phenomenal consciousness and different settings of attention, trained meditators (e.g. Buddhist meditation practitioners) with a long-term mental practice could be usefully involved in such investigations (Lutz, Slagter, Dunne, & Davidson, 2008; see also Raffone & Srinivasan, 2009, 2010). Acknowledgments We would like to thank Dr. Bridgeman and three anonymous reviewers of ‘‘Consciousness and Cognition” for their extremely useful remarks and comments. Appendix A The neural activation level xi of fast excitatory units is computed by the following equation:
dxi ¼ Afast xi þ ðB xi ÞðCfa ðxi Þ þ ExcVI þ ExcVD þ gÞ xi ðDfa ðyi Þ þ InhÞ dt
ð1Þ
In Eq. (1) Afast is the decay parameter, the term ðB xi Þ is the shunting term that automatically controls the excitatory inputs. The inhibitory inputs (see the last term) are controlled via multiplication by xi. C is the self-excitation strength, fa is a sigmoid function (see below). ExcVI the voltage independent excitatory input, ExcVD the voltage-dependent excitatory input, Inh is the inhibitory input (see below) associated to competing objects. D is the strength of inhibition from the coupled inhibitory unit. The symbol g denotes uniformly distributed random noise. The activation level yi of the coupled inhibitory unit, is computed by the following equation:
dyi ¼ Eðyi xi Þ dt
ð2Þ
In Eq. (2) E is the linear gain/decay parameter. The activation level of slow excitatory units zi, is computed via the following equation:
dzi ¼ Aslow zi þ ðB zi Þcfast fb ðxi Þ dt
ð3Þ
In Eq. (3) Aslow is a decay parameter, (B zi) is a shunting term for excitatory inputs, cfast is the connection strength from the coupled fast excitatory unit. In the GW gate layer, a slowly rising neuronal adaptation current ui is computed, via the following equation:
dui ¼ Fðzi ui Þ dt
ð4Þ
In Eq. (4) F is a linear gain/decay term. The ui value is an additional inhibitory input for fast excitatory units in the GW gate (see Eq. (1)). To model inter-object competition in all layers except for the sensory map, the competitive inhibitory activation level vi is computed, via the following equation:
dv i ¼ Acomp v i þ ðB v i Þxi dt
ð5Þ
where Acomp is a decay term, and (B vi) a shunting term of the excitatory input from the coupled fast excitatory unit. The voltage-independent excitatory inputs to the fast excitatory units ExCVI (see Eq. (1)) are computed as follows:
ExcVI ¼ cstim Istim þ cslow fb ðzi Þ þ clat
N X
fb ðxj Þ þ cforw xlow i
ð6Þ
j¼1ði–jÞ
where Istim is a binary external input multiplied by a strength parameter cstim, cslow fb (zi) is the input term by the coupled PN j¼1ði–jÞ fb ðxj Þ is the input term from the set of coupled fast excitatory units coding for the same object
slow excitatory unit, clat
is the input term from a paired unit in an earlier layer (forward input). in the same layer, cforw xlow i The voltage-dependent excitatory inputs to the fast excitatory units ExcVD (see Eq. (1)) are computed as follows:
ExcVD ¼ ðcfback zhigh þ Iatt Þ i ExcVD ¼ 0 ifxi 6 VT
xi VT 1 VT
if xi > VT
ð7aÞ ð7bÞ
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
595
where cfback zhigh is the (feedback) input from a paired slow excitatory unit in the upper layer, Iatt is a top-down attentional i i VT input from implicitly modeled areas, and the terms x1VT is a scaling term. VT is a voltage-threshold (see Tononi et al. (1992), for a similar computation of voltage-dependent inputs). The competitive inhibitory net input is computed via the following equation:
Inh ¼ ccomp
Nc X
f ðv j Þ
ð8Þ
j¼1
as a weighted sum over all the activations of inhibitory units for competition, each paired to a fast excitatory unit for another object (see Eq. (5)). The generic sigmoid function f(x) with parameters n and Q, is computed according to the following formula (see also Grossberg & Grunewald, 1997):
FðxÞ ¼
xn Q n þ xn
ð9Þ
The parameters in the equations above used in the simulations were set as follows: Afast = 1; B = 1; C = 20; D = 33.3; E = 0.2; Qa = .9; na = 4 Aslow = 0.01; cfast = 0.2; F = 0.003; Acomp = 0.1; cstim, = 0.3; cslow = 0.175; clat = 0.015; cf orw = 0.375; cfback = 2; Iatt = 0.3 at the phenomenal register layer, and Iatt = 1 at the consumer systems layer; VT = 0.1; Qb = 0.6; nb = 4; ccomp = 0.075 at the GW gate layer, and ccomp = 0.01 at the phenomenal register and consumer systems layers; the noise term g was uniformly distributed in the range [0, 0.05]. The initial values of the neural state variables were set equal to 0, except for xt, chosen randomly in the range [0, 0.15], and yi, chosen randomly in the range [0, 0.4]. Numerical integration was performed with the Euler method, with step equal to 0.01. Correspondence to real time was set with 50 numerical integration steps for 1 ms of real time. References Baars, B. J. (1983). Conscious contents provide the nervous system with coherent, global information. In R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.), Consciousness & self-regulation (p. 41). NY: Plenum Press. Baars, B. (1998). Metaphors of consciousness and attention in the brain. Trends in Neurosciences, 21, 58–62. Baars, B. J., Ramsoy, T. Z., & Laureys, S. (2003). Brain, conscious experience and the observing self. Trends in Neuroscience, 26, 671–675. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Science, 4, 417–423. Baddeley, A. D., & Hitch, G. J. L. (1974). Working memory. In G. A. Bower (Ed.). The psychology of learning and motivation: Advances in research and theory (Vol. 8, pp. 47–89). New York: Academic Press. Behrmann, M., Geng, J. J., & Shomstein, S. (2004). Parietal cortex and attention. Current Opinion in Neurobiology, 14, 212–217. Blake, R., & Logothetis, N. K. (2002). Visual competition. Nature Review Neuroscience, 3, 13–21. Block, N. (1995). On a confusion about a function of consciousness. Behavioural and Brain Sciences, 18, 227–287. Block, N. (2005). Two neural correlates of consciousness. Trends in Cognitive Sciences, 9, 46–52. Block, N. (2007). Consciousness, accessibility, and the mesh between psychology and neuroscience. Behavioural and Brains Sciences, 30, 481–548. Brunel, N., & Wang, X. J. (2001). Effects of neuromodulation in a cortical network model of object working memory dominated by recurrent inhibition. Journal of Computational Neuroscience, 11, 63–85. Buehlmann, A., & Deco, G. (2008). The neuronal basis of attention: Rate versus synchronization modulation. Journal of Neuroscience, 28, 7679–7768. Bundesen, C. (1990). A theory of visual attention. Psychological Review, 97, 523–547. Bundesen, C., Habekost, T., & Kyllingsbaek, S. (2005). A neural theory of visual attention: Bridging cognition and neurophysiology. Psychological Review, 112, 291–328. Cartwright-Finch, U., & Lavie, N. (2006). The role of perceptual load in inattentional blindness. Cognition, 102, 321–340. Chun, M. M., & Potter, M. C. (1995). A two-stage model for multiple target detection in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 21, 109–127. Cohen, J. D., Braver, T. S., & Brown, J. W. (2002). Computational perspectives on dopamine function in prefrontal cortex. Current Opinion in Neurobiology, 12, 223–229. Colzato, L. S., Slagter, H. A., Spapè, M., & Hommel, B. (2008). Blinks of the eye predict blinks of the mind. Neuropsychologia, 46, 3179–3183. Corchs, S., & Deco, G. (2004). Feature-based attention in human visual cortex: Simulation of fMRI data. NeuroImage, 21, 36–45. Dehaene, S., & Changeux, J. P. (2005). Ongoing spontaneous activity controls access to consciousness: A neuronal model for inattentional blindness. PLoS Biology, 3(5), e141. Dehaene, S., Changeux, J. P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: A testable taxonomy. Trends in Cognitive Sciences, 10, 204–211. Dehaene, S., Kerszberg, M., & Changeux, J. P. (1998). A neuronal model of a global workspace in effortful cognitive tasks. Proceedings of the National Academy of Sciences USA, 95, 14529–14534. Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79, 1–37. Dehaene, S., Sergent, C., & Changeux, J.-P. (2003). A neuronal network model linking subjective reports and objective physiological data during conscious perception. Proceedings of the National Academy of Science USA, 100, 8520–8525. Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective visual attention. Annual Review of Neuroscience, 18, 193–222. Duncan, J. (1984). Selective attention and the organization of visual information. Journal of Experimental Psychology: General, 114, 501–517. Duncan, J. (2001). An adaptive coding model of neural function in prefrontal cortex. Nature Reviews Neuroscience, 2, 820–829. Duncan, J., & Miller, E. K. (2002). Cognitive focusing through adaptive neural coding in the primate prefrontal cortex. In D. Stuss & R. T. Knight (Eds.), Principles of frontal lobe function (pp. 278–291). Oxford University Press. Edelman, G., & Tononi, G. (2000). Consciousness. How matter becomes imagination. London: Penguin Books. Fries, P., Reynolds, J. H., Rorie, A. E., & Desimone, R. (2001). Modulation of oscillatory neuronal synchronization by selective visual attention. Science. Gaillard, R., Dehaene, S., Adam, C., Clémenceau, S., Hasboun, D., et al (2009). Converging intracranial markers of conscious access. PLoS Biology, 7(3), e1000061. Grossberg, S., & Grunewald, A. (1997). Cortical synchronization and perceptual framing. Journal of Cognitive Neuroscience, 9, 117–132.
596
A. Raffone, M. Pantani / Consciousness and Cognition 19 (2010) 580–596
Grossberg, S., & Somers, D. (1991). Synchronized oscillations during cooperative feature linking in a cortical model of visual perception. Neural Networks, 4, 453–466. Lamme, V. (2003). Why attention and awareness are different. Trends in Cognitive Sciences, 7, 12–18. Lamme, V. (2004). Separate neural definitions of visual consciousness and visual attention; a case for phenomenal awareness. Neural Networks, 17, 861–872. Lamme, V. A. F., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences, 23, 571–579. Landman, R., Spekreijse, H., & Lamme, V. A. F. (2003). Large capacity storage of integrated objects before change blindness. Vision Research, 43, 149–164. Lumer, E. D., Edelman, G. M., & Tononi, G. (1997a). Neural dynamics in a model of the thalamocortical system. 1. Layers, loops and the emergence of fast synchronous rhythms. Cerebral Cortex, 7, 207–227. Lumer, E. D., Edelman, G. M., & Tononi, G. (1997b). Neural dynamics in a model of the thalamocortical system. 2. The role of neural synchrony tested through perturbations of spike timing. Cerebral Cortex, 7, 228–236. Luria, A. (1974). The working brain. New York: Basic Books. Lutz, A., Slagter, H. A., Dunne, J. D., & Davidson, R. J. (2008). Attention regulation and monitoring in meditation. Trends in Cognitive Sciences, 12, 163–169. Lutz, A., & Thompson, E. (2003). Neurophenomenology. Journal of Consciousness Studies, 10, 31–52. McClelland, J. L., & Rumelhart, D. E. (1986). Parallel distributed processing: Explorations in the microstructure of cognition (Vol. II). Cambridge, MA: MIT Press. Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24, 167–202. Murray, S. O., & Wojciulik, E. (2004). Attention increases neural selectivity in the human lateral occipital complex. Nature Neuroscience, 7, 70–74. Overgaard, M., Rote, J., Mouridsen, K., & Ramsøy, T. Z. (2006). Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task. Consciousness and Cognition, 15, 700–708. Owen, A. M., Evans, A. C., & Petrides, M. (1996). Evidence for a two-stage model of spatial working memory processing within the lateral frontal cortex: A positron emission tomography study. Cerebral Cortex, 6, 31–38. Pashler, H., & Johnston, J. C. (1998). Attentional limitations in dual-task performance. In H. Pashler (Ed.), Attention (pp. 155–189). Hove, England UK: Psychology Press/Erlbaum (UK) Taylor & Francis. Posner, M. I., & Fan, J. (2004). Attention as an organ system. In J. R. Pomerantz & M. C. Crair (Eds.), Topics in integrative neuroscience: From cells to cognition. Cambridge UK: Cambridge University Press. Raffone, A., & Srinivasan, N. (2009). An adaptive workspace hypothesis about the neural correlates of consciousness: Insights from neuroscience and meditation studies. In N. Srinivasan (Ed.). Progress in brain research: Attention (Vol. 176, pp. 161–180). Amsterdam: Elsevier (pp. 171–180. Raffone, A., & Srinivasan, N. (2010). The exploration of meditation in the neuroscience of meditation and consciousness. Cognitive Processing, 11, 1–7. Raffone, A., & Wolters, G. (2001). A cortical mechanism for binding in visual working memory. Journal of Cognitive Neuroscience, 13, 766–785. Raymond, J. E., Shapiro, K. L., & Arnell, K. M. (1992). Temporary suppression of visual processing in an RSVP task: An attentional blink? Journal of Experimental Psychology, Human Perception and Performance, 18, 849–860. Sergent, C., & Dehaene, S. (2004). Is consciousness a gradual phenomenon? Evidence for an all-or-none bifurcation during the attentional blink. Psychological Science, 15, 720–729. Sligte, I. G., Scholte, H. S., & Lamme, V. (2008). Are there multiple visual short-term memory stores? Plos, 3, e1699. Sperling, G. (1960). The information available in brief visual presentations. Psychological Monographs: General and Applied, 74, 1–30. Srinivasan, N., Srivastava, P., Lohani, M., & Baijal, S. (2009). Focused and distributed attention. In N. Srinivasan (Ed.), Progress in brain research: Attention. Vol. 176 (pp. 87–100). Amsterdam: Elsevier. Tononi, G., Sporns, O., & Edelman, G. M. (1992). Reentry and the problem of integrating multiple cortical areas: Simulation of dynamic integration in the visual system. Cerebral Cortex, 2, 310–335. Treisman, A. (2006). How the deployment of attention determines what we see. Visual Cognition, 14, 411–443. Usher, M., & Niebur, E. (1996). A neural model for parallel, expectation-driven attention for objects. Journal of Cognitive Neuroscience, 8, 305–321. VanRullen, R., & Koch, C. (2003). Competition and selection during visual processing of natural scenes and objects. Journal of Vision, 3, 75–85. Varela, F. (1996). Neurophenomenology: A methodological remedy for the hard problem. Journal of Consciousness Studies, 3, 330–349. Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal individual differences in controlling access to working memory. Nature, 438, 500–503. Wager, T. D., & Smith, E. E. (2003). Neuroimaging and working memory: A meta-analysis. Cognitive, Affective, and Behavioral Neuroscience, 3, 255–274. Wang, X. J. (1999). Synaptic basis of cortical persistent activity: The importance of NMDA receptors to working memory. Journal of Neuroscience, 19, 9587–9603. Wang, Y., Markram, H., Goodman, P. H., Berger, T. K., Ma, J., & Goldman-Rakic, P. S. (2006). Heterogeneity in the pyramidal network of the medial prefrontal cortex. Nature Neuroscience, 9, 534–542.
Consciousness and Cognition 19 (2010) 597–605
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Mindfulness meditation improves cognition: Evidence of brief mental training q Fadel Zeidan a,*, Susan K. Johnson b, Bruce J. Diamond c, Zhanna David b, Paula Goolkasian b a
Department of Neurobiology and Anatomy, Wake Forest University School of Medicine, USA Department of Psychology, University of North Carolina, Charlotte, USA c Department of Psychology, William Patterson University, USA b
a r t i c l e
i n f o
Article history: Received 16 December 2009 Available online 3 April 2010 Keywords: Mindfulness Meditation Cognition Working memory Mood Attention Meta-awareness
a b s t r a c t Although research has found that long-term mindfulness meditation practice promotes executive functioning and the ability to sustain attention, the effects of brief mindfulness meditation training have not been fully explored. We examined whether brief meditation training affects cognition and mood when compared to an active control group. After four sessions of either meditation training or listening to a recorded book, participants with no prior meditation experience were assessed with measures of mood, verbal fluency, visual coding, and working memory. Both interventions were effective at improving mood but only brief meditation training reduced fatigue, anxiety, and increased mindfulness. Moreover, brief mindfulness training significantly improved visuo-spatial processing, working memory, and executive functioning. Our findings suggest that 4 days of meditation training can enhance the ability to sustain attention; benefits that have previously been reported with long-term meditators. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction People who have undergone extensive meditation training have shown improvements on cognitive performance (Cahn & Polich, 2006) and mood (Davidson et al., 2003). Long-term meditation practice has been found to enhance attentional (Jha, Krompinger, & Baime, 2007) and visuospatial processes (Kozhevnikov, Louchakova, Josipovic, & Motes, 2009). For example, 3-months of intensive meditation training (10–12 h/day) improved the ability to sustain attention during a dichotic listening task as evidenced by faster reaction times in response to a deviant tone, and reduced attentional blink responses when compared to controls (Lutz et al., 2009; Slagter, Lutz, Greischer, Nieuwenhuis, & Davidson, 2009; respectively). Moore and Malinowski (2009) found that self-reported mindfulness was positively correlated with sustained attention in experienced Buddhist meditation practitioners, when compared to controls. Additionally, long-term meditation practice has been found to reduce attentional blink in older adults when compared to age-matched and younger adults (van Leeuwen, Muller, & Melloni, 2009). In a study employing neuroimaging (Short et al., 2007), extensive meditation training heightened activation in executive attention networks that was correlated with improvements in sustained attention and error monitoring. These findings provide growing evidence of mindfulness meditation’s (MM) promotion of higher-order cognitive processing; specifically facets of conflict monitoring and cognitive control processes. Mindfulness Based Stress Reduction (MBSR) programs, which are usually 8 weeks in duration and combine mindfulness meditation and gentle yoga, have been found to improve mood and affective processes (Nyklícek & Kuijpers, 2008); and are q
This research was part of the doctoral dissertation of the first author under the supervision of the second author. * Corresponding author at: Wake Forest University School of Medicine, Winston-Salem, NC 27157, USA. E-mail address:
[email protected] (F. Zeidan).
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.014
598
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
associated with improvements in immune system functioning (Davidson et al., 2003), stress (Carlson, Speca, Faris, & Patel, 2007), and emotional regulation (Nielsen & Kaszniak, 2006). MBSR programs are based on teaching participants to react nonjudgmentally to stressful events by focusing on automatic and dynamic stimuli (breath; body; eating; walking). As participants cultivate these skills, top-down control processes regulate affective appraisals that lead to a reduction in stress responses (Grossman, Niemann, Schmidt, & Walach, 2004). In an elegant study, a MBSR program promoted decreases in stress ratings which were correlated with reductions in amygdala gray matter density; providing objective evidence for the positive effects of MBSR on stress (Hölzel et al., 2009). Although advantageous to well-being, MBSR programs require extensive time and financial commitment. Consequently, most individuals do not have the time or resources to participate in extensive meditation interventions and few will choose the monastic lifestyle that is often associated with Buddhist contemplatives. Studying adept meditators is invaluable to understanding the aptitude of human consciousness, however, it is important to investigate whether briefer formats of mental training can provide some of the benefits that result from longer interventions. MM is a mental practice based on focusing on the sensations of the breath/body while maintaining a relaxed state of mind. During formal meditation practice, distractions will arise and the meditator is taught to acknowledge discursive thoughts, and non-judgmentally return his/her attention back to their breathing (Wallace, 2006). Mindfulness training cultivates moment-to-moment awareness of the self and environment (Wallace, 2006). To this extent, mindfulness training heightens meta-cognitive processing (Austin, 1998). Meta-cognition is the conscious awareness of cognitive control processes (Fernandez-Duque, Baird, & Posner, 2000). Improvements in meta-cognition are related to the ability to restrict bottom-up processing of exogenously/endogenously driven, task-irrelevant information (Posner & Rothbart, 1998). Extensive training in mindfulness has been found to improve alerting and conflict monitoring Cahn & Polich, 2006; Jha et al., 2007), therefore mindfulness meditation training can hypothetically enhance meta-awareness. This process improves attention sustainability by teaching subjects to ‘‘release” cognitive appraisals of irrelevant information. So far, however, the cognitive benefits associated with mindfulness have been limited to studies examining adept meditators (Cahn & Polich, 2006). Although research examining the effects of extensive meditation interventions is growing, the effects of brief mental training on mood and cognition are relatively unknown. We examined whether 4 days (20 min/day) of MM training affects behavioral markers of cognition and mood. Tang et al. (2007) reported that 5 days of Integrative Body Mind Training improved mood and cognitive processes. However, Integrative Body Mind Training incorporates various techniques (e.g. mindfulness, guided-imagery, music therapy) leaving it hard to decipher if mindfulness was the mechanism underlying improvements. A recent study (Zeidan, Gordon, & Goolkasian, 2009) found that 3 days of MM training was effective at reducing pain ratings and sensitivity, as well as anxiety scores when compared to baseline and other cognitive manipulations, such as relaxation and a math distracter task. A similar training regimen improved mood and reduced heart rate when compared to a sham MM and control group (Zeidan, Johnson, Gordon, & Goolkasian, in press). The present study builds on our previous work by examining if the effects of brief meditation training can be found on cognitive tasks with varying demands on working memory, sustained attention, visual coding, and verbal fluency. We used the Symbol Digit Modalities Test and the n-back task to measure working memory, processing speed, and executive attention; the forward and backward digit span to measure immediate attention span, and the Controlled Oral Word Association Test to measure verbal fluency. Based on prior research (Cahn & Polich, 2006), we expected that these tasks would be the most sensitive to the effects of meditation. We also expected that brief MM training, when compared to a group that listened to a book recording, would promote positive mood, as measured by the Center for Epidemiologic Studies Depression scale (CES-D), State Anxiety Inventory, and the Profile of Mood States. 2. Methods 2.1. Participants Sixty-three University of North Carolina, Charlotte students volunteered for the experiment in fulfillment of General Psychology requirements. The participants were recruited from those who were interested in learning meditation and who had no prior meditation experience. The consent form explained that subjects would be randomly assigned to a meditation or a book listening group. Group assignment depended upon the week that the participant signed up for the study. Seven participants from each of the groups did not complete the protocol and their data were not included. From the remaining participants, 24 were assigned to the meditation group and 25 to the control. Table 1 compares the groups on demographic (age; gender; ethnicity) and baseline measures and shows that there were no differences. The median age was 20 years. Sixty-one percent of the participants were White, 25% were African–American, 2% were Asian, and 4% were biracial, Native American, and Hispanic. 2.2. Interventions 2.2.1. Mindfulness meditation Mindfulness training was modeled on basic Shamatha skills (Wallace, 2006). Meditation training was conducted by a facilitator with 10 years experience in teaching MM. In session one, small groups of three to five participants were instructed
599
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605 Table 1 Mean (SD) scores for each group on baseline measures. Meditation
Controls
t/v2*
P
Age Caucasian Female
22 (7.90) 67% 63%
23 (8.36) 56% 56%
.66 7.12* .21*
.51 .21 .64
Self-report measures Freiburg Mindfulness Inventory State Anxiety Inventory POMS CES-D
43.33 36.75 15.21 15.21
47.00 35.28 12.00 13.08
1.53 .51 .42 .81
.13 .61 .68 .42
Cognitive measures SDMT Forward digit span Backward digit span COWAT Extended hit rate Accuracy proportion Processing speeda
58.79 (10.78) 10.88 (2.58) 6.58 (2.60) 35.92 (8.61) 2.38 (3.41) .70 (.07) 130.03 (139.76)
.24 .05 .21 .31 1.33 .73 .63
.81 .96 .84 .76 .19 .47 .53
(8.57) (10.72) (26.39) (7.82)
(8.24) (9.59) (27.49) (10.13)
59.52 (10.51) 10.84 (2.15) 6.72 (1.99) 36.72 (9.74) 3.48 (4.58) .68 (.13) 104.74 (116.74)
df = 1,47. (SD) Standard deviations are given in parentheses. POMS (Total Profile Mood State). CES-D (Center for Epidemoplogic Studies Depression Scale), SDMT (Symbol Digit Modalities Test), COWAT (Controlled Oral Word Association). a Processing speed measured in seconds. * p < .05.
to relax, with their eyes closed, and to simply focus on the flow of their breath occurring at tip of their nose. If a random thought arose, they were told to passively notice and acknowledge the thought and to simply let ‘‘it” go, by bringing the attention back to the sensations of the breath. In subsequent sessions (2–4), participants worked on developing mindfulness skills. For example, in sessions 2 and 3, subjects were taught to focus on the full breath, that is, to focus on the sensations of the breath from the nostrils to the abdomen and back. Participants were also taught to notice and focus on any sensations that arose in the body, and to simply acknowledge those feelings and then to return their attention back to their breath (Wallace, 2006). Session 4 was premised on developing the skills established in the previous sessions, however, more time was spent in silence so that participants could meditate. As a manipulation check, each subject was asked after each meditation session, ‘‘Did you feel that you were truly meditating?” Across all sessions, participants were consistent in acknowledging that they were truly meditating. 2.2.2. Control group Control participants were instructed to listen in small groups to JRR Tolkein’s The Hobbit on compact disc (BBC audiobooks Ltd., 1997). The beginning of the story was played in session 1 and the following sessions (2–4) continued with the story. They were instructed to silence cell phones and any electronics, sit quietly and listen to the audio book. A research assistant sat with the participants to monitor attentiveness during the listening task. 3. Materials 3.1. Self-report measures We administered the Freiburg Mindfulness Inventory, State Anxiety Inventory, CES-D, and Profile of Mood States on session 1 before the interventions and on session 4 after the interventions. The Profile of Mood States (McNair, Loor, & Droppleman, 1971) is a 65-item inventory that measures current mood state by rating adjective like statements (e.g. I feel calm) on a Likert scale (0–4). It consists of six subscales: tension, depression, confusion, fatigue, anger, and vigor: and a total negative mood score is calculated by subtracting the vigor scale from the sum of the remaining subscales. The CES-D (Radloff, 1997) is a well-validated, 20-item, scale that measures depressive symptomatology. Subjects were asked to rate statements, based on the ‘‘last week,” such as ‘‘I felt happy” and ‘‘I felt depressed” on a four point Likert scale (0–3) and scores range from 0 to 60 with higher scores indicating higher levels of depression. The Freiburg Mindfulness Inventory is a 14-item assessment that measures the experience of mindfulness (Walach, Buchheld, Buttenmuller, Kleinknecht, & Schmidt, 2006). It is a psychometrically valid instrument with high internal consistency (Cronbach alpha = . 93) (Baer, Smith, Hopkins, Krietemeyer, & Toney, 2007). Statements like ‘‘I am open to the experience of the present moment,” are rated on a four-point scale from 1 (rarely) to 4 (always). Scores range from 14 to 56 and higher scores indicate a greater degree of mindfulness. The Freiburg Mindfulness Inventory served as a manipulation check on participants’ ability to engage in a ‘‘mindful” state, and was also used to measure changes in mindfulness that resulted from the training. The State Anxiety Inventory is a 20-item scale designed to measure state anxiety (Spielberger, 1983). It has high internal consistency with Cronbach’s alpha of .73 (Spielberger, 1983). Statements like ‘‘I feel worried,” are rated on a four-point scale from 1 (not at all) to 4 (very much so). Scores range from 20 to 80 and higher scores indicate higher levels of anxiety.
600
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
3.2. Cognitive measures Standardized cognitive tasks, as well as a computer adaptive n-back task, were administered before (session 1) and after (session 4) the intervention. Alternate versions of all tasks (except the Symbol Digit Modalities Test) were used across sessions. The Controlled Oral Word Association Test (Benton, 1989) is a measure of verbal fluency in which subjects are asked to say as many words as they can think of beginning with the letters ‘‘F, A, and S”, or ‘‘C, F, and L” within one minute. The dependent measure is the total number of words produced. The Symbol Digit Modalities Test (Smith, 1982), written version, is a measure of complex visual tracking and working memory that requires decoding of a series of numbers listed on paper according to a corresponding template of visual symbols. With the use of a reference key, participants were given 90 s to accurately match numbers with corresponding geometric figures. The dependent measure is number of symbols coded minus errors. The forward/backward digit span (Wechsler Adult Intelligence Scale-Revised (WAIS-R) (Wechsler, 1981) was used to measure immediate memory span. The dependent measures are total forward digit span and total backward digit span. In the forward digit span, subjects can correctly repeat back a span of up to16 digits. In the backward version, subjects can recite back a span of up to 14 digits backwards. Higher scores are indicative of higher memory recall. The computer adaptive n-back task is an adaptive, accuracy selectable, 2-back task that consists of 54 trials and was developed in order to measure information processing speed, working memory and attention. It represents an advancement over previously used n-back tasks (as cited in Strauss, Sherman, & Spreen, 2006) because it corrects for the accuracy-speed confound by allowing the experimenter to set desired accuracy levels and by equating accuracy levels across groups. The program’s algorithms (V2ASQQQQ) variably adjust the presentation speed based on individual trial responses, response patterns and multiple accuracy windows to a value that supports the desired accuracy. Participants view a sequence of letters and indicate whether or not a probe letter is the ‘‘same” or ‘‘different” as the stimulus item presented two items back. The program computes two measures – speed of processing and ‘‘extended hit rate” (representing a run of correct responses). A measure of accuracy was also computed to make sure that there were functionally equivalent levels across groups.
4. Procedure Participants were assigned to intervention groups based on the particular week that they enrolled in the experiment. Mindfulness meditation and control group interventions were randomly assigned to weeks. The interventions met on the same days of the week at the same time of day. 4.1. Session 1 After obtaining consent forms, participants completed the Freiburg Mindfulness Inventory, the Profile of Mood States, the State Anxiety Inventory, CES-D in random order followed by the Symbol Digit Modalities Test, backwards/forward digit span recall test, the Controlled Oral Word Association Test, and computer adaptive two-back task in random order. Based on group assignment, subjects were either led in meditation or listened to The Hobbit in a small group setting for 20 min in each session. Afterwards, the State Anxiety Inventory was reassessed. 4.2. Sessions 2–3 Depending on group assignment, participants either came in for meditation training or book listening, and the State Anxiety Inventory was completed before and after the meditation/listening intervention. 4.3. Session 4 Participants in the meditation and listening group completed the State Anxiety Inventory before their respective intervention. At the completion of the intervention, the Freiburg Mindfulness Inventory, State Anxiety Inventory, Profile of Mood States, CES-D, and the cognitive tasks were administered. In general, the statistical analyses tested for the between group effect of intervention training and the within group effect of pre/post (session 1 vs. session 4) intervention training. The State Anxiety Inventory was administered before and after the intervention in each session. Multivariate analyses of variance (MANOVA) were used (with Wilk’s criterion as the test statistic) on the subscale scores of the Profile of Mood States, and the scores from the cognitive tasks (Symbol Digit Modalities Test, Fluency, Backward/Forward recall tests, extreme hit rate, and speed of processing). Follow-up univariate analyses were conducted when appropriate. Total Profile of Mood States, State Anxiety Inventory, CES-D, and Freiburg Mindfulness Inventory were analyzed with separate mixed analysis of variances (ANOVA). A significance level of .05 was used for all statistical tests.
601
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
5. Results 5.1. Self-report measures Table 2 reports the Freiburg Mindfulness Inventory and Profile of Mood States data for each group across sessions. The analysis on the Freiburg Mindfulness Inventory scores showed that brief MM training was effective at increasing mindfulness skills when compared to controls. There was a group by session interaction, F(1, 47) = 5.73, p = .02, g2 = .11, and an effect of session, F(1, 47) = 13.18, p = .001, g2 = .22; but no effect of group, F < 1. The analysis on the total scores from the Profile of Mood States indicated a reduction in negative mood across session, F(1, 47) = 19.50, g2 = .29; however, there were no group differences, F < 1, and group did not interact with session, F < 1. The MANOVA on the six subscales of the Profile of Mood States was similar in showing a session effect, F(6, 42) = 4.50, p = .001, g2 = .39; and there was no main effect for group, F(6, 42) = 1.94, p = .10; but there was a marginally significant group by session interaction, F(6, 42) = 2.22 p = .06, g2 = .24. Session effects on all of the subscales except vigor, F(1, 47) = 3.27, p = .08, showed that both groups improved from session 1 to session 4 (fatigue, F(1, 47) = 5.26, p = .03, g2 = .10; depression F(1, 47) = 13.31, p = .001, g2 = .22; tension ratings, F(1, 47) = 27.79, p < .001, g2 = .37; anger, F(1, 47) = 10.61, p = .002, g2 = .18; confusion, F(1, 47) = 7.35, p = .009, g2 = .14). Moreover, the analysis on the fatigue subscale showed that the session effect varied by group, F(4, 47) = 5.05, p = .03, g2 = .10. Brief meditation training was effective at significantly reducing fatigue, while listening to the book did not. However, some caution must be exercised because of the marginal significance of the interaction effect in the MANOVA. The group by session interaction was not present in the analyses of the other subscales. State Anxiety Inventory scores, presented in Table 3, were analyzed with a two (group) two (before/after) four (session) ANOVA. As expected, the anxiety scores dropped significantly after practice with meditation but not after listening to the book. The analysis indicated a before/after effect, F(1, 47) = 110.03, p < .001, g2 = .70, and before/after interacted with group, F(1, 47) = 40.19, p < .001, g2 = .46. There was also a decline in state anxiety scores across session, F(3, 141) = 6.85, p < .001, g2 = .13. However, there were no other interactions; session did not interact with group, (F < 1), or in a threeway interaction with group, and before/after, F(3, 141) = 1.47, p = .23. The analysis of the CES-D scores were similar to the Profile of Mood States subscale for depression in showing a significant decrease in depression scores from session 1(M = 14.12, SD = 10.08) to 4 (M = 9.28, SD = 5.13), F(1, 47) = 8.69, p = .005, g2 = .16; but there were no effects of group and no interaction between session and group (Fs < 1). 5.2. Cognitive tasks Fig. 1 shows that brief mindfulness training was effective at improving performance on several cognitive tasks—Symbol Digit Modalities Test, verbal fluency, and the n-back task. The MANOVA on the scores from the Symbol Digit Modalities Test, verbal fluency, the forward and backward digit span, and the two measures of the n-back task showed a significant group by session interaction, F(6, 42) = 2.28, p = .05, g2 = .25; a main effect of session, F(6, 42) = 10.66, p < .001, g2 = .60; and no effect of group, F < 1. Follow-up ANOVAs on symbol digit modality and verbal fluency showed a significant improvement in performance across sessions for the meditation group but not for the control group. The two analyses showed significant interaction effects: Symbol Digit Modalities Test, F(1, 47) = 6.78, p = .01, g2 = .13, verbal fluency, F(1, 47) = 5.27, p = .03, g2 = .10, and main effects of session, Symbol Digit Modalities Test, F(1, 47) = 18.78, p < .001, g2 = .29 and verbal fluency, F(1, 47) = 12.45, p = .001, g2 = .21. Performance on both the forward and backward digit span improved after training but the change was not specific to the MM group. There was a session effect, (forward, F(1, 47) = 12.63, p = .001, g2 = .21; backward, F(1, 47) = 12.62, p = .01,
Table 2 Mean (SD) for each group on POMS and FMI. Meditationa Pre FMI Total POMS Tension Depression Anger Vigor Fatigue Confusion
43.33 15.20 8.50 6.13 4.00 15.88 5.54 6.92
Controlb Post
(8.57) (26.39) (6.95) (7.94) (5.89) (5.49) (4.33) (3.88)
Total (total mood score), (Freiburg Mindfulness Inventory). a n = 24. b n = 25.
49.96 4.80 2.91 1.95 1.91 14.04 1.71 5.46
Pre (10.0) (15.63) (2.75) (3.99) (2.88) (5.85) (3.17) (3.56)
47.00 12.00 6.28 4.52 3.80 18.75 4.84 6.60
Post (8.24) (27.49) (6.14) (8.38) (5.73) (7.17) (5.32) (3.72)
48.36 1.16 2.60 .52 1.20 15.16 4.80 4.88
(8.42) (11.78) (2.22) (1.16) (1.98) (6.33) (5.57) (2.20)
602
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
Table 3 Means (SD) for control and meditation group on State Anxiety Inventory scores across sessions. Meditationa
Session Session Session Session
1 2 3 4
M (SD) a b
Controlb
Before
After
36.75 33.75 32.67 32.29
27.58 24.96 25.29 25.67
(10.72) (9.08) (7.33) (8.30)
33.86 (5.98)
(5.76) (5.34) (4.15) (5.28)
25.88 (3.67)
Before
After
35.28 30.48 31.68 30.40
35.12 27.24 29.32 28.28
(9.33) (6.32) (7.64) (7.54)
31.96 (4.46)
(10.51) (5.91) (7.83) (6.17)
29.99 (5.53)
n = 24. n = 25.
Fig. 1. Mean scores for the group by session interaction from the Symbol Digit Modalities test, Controlled Oral Word Association Test, and the Computer Adaptive 2-back task. Error bars represent 95% confidence intervals.
g2 = .13), but session did not interact with group, (forward F(1, 47) = 1.27, p = .27; backward F(1, 47) = 1.26, p = .27). Means scores on the forward digit span task for sessions 1 and 4 are as follows, 10.85 and 12.02. Means scores on the backward digit span task for sessions 1 and 4 are as follows, 6.60 and 7.10. The ANOVA on the accuracy measure from the computer adaptive 2-back task found no differences across sessions or between groups (Fs < 1), indicating that the 2-back task was performed at the same level of accuracy across the two conditions. Fig. 1 shows the significant group by session interaction found in the analysis on the extended hit rate from the computer adapted n-back task, F(1, 47) = 6.76, p = .01, g2 = .12. The meditation group, in contrast to the control group, had more extended hit runs. There was also a significant session effect, F(1, 47) = 18.78, p < .001, g2 = .29. Follow-up analyses on the speed measure of the n-back task did not show any evidence of an interaction effect F < 1, or an effect of session or group, Fs < 1. 6. Discussion Our findings, with naïve participants learning mindfulness techniques by means of a brief training format, are consistent with those that have been reported for adept meditators. Four days (20 min/day) of MM training was effective in significantly increasing mindfulness scores in comparison to an active control group. Our brief MM training protocol promoted significant effects on several cognitive tasks that require sustained attention and executive processing efficiency (Symbol Digit Modalities Test, verbal fluency, and the hit runs on n-back task). However, no specific benefits from MM training were found with many of the mood scales, the forward and backward digit span or speed on the n-back task. It is important to note that the groups did not differ at baseline on any of the measures and had no prior meditative experience.
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
603
A previous study (Zeidan et al., in press) that examined brief MM training, found more pervasive mood changes when comparing MM to sham meditation and control interventions. In the current study, listening to the book was intended to occupy their attention and not expected to affect mood. However, it may have served as a relaxing activity, which could explain the decrease in negative mood after book listening. In a recent review examining the efficacy of mindfulness interventions, MBSR programs were not more effective than active control groups (e.g. relaxation; cognitive therapy) on mood outcomes, when compared to wait-listed groups (Toneatto & Nguyen, 2007). There are a number of possible mechanisms that may explain the relation between mindfulness and cognitive improvement. In contrast to controls, brief MM training reduced participants’ fatigue and anxiety ratings. Fatigue and anxiety may be particularly critical in affecting information processing. More participants would be needed, however, to properly assess the relatedness among these measures. Prior studies with multiple sclerosis patients have found that fatigue adversely affects complex visual tracking speed measured by the Symbol Digit Modalities Test, (Andreasen, Spliid, Andersen, & Jakobsen, 2009), 2-back task processing (Diamond, Johnson, Kaufman, & Graves, 2008), and sustained attention on the Paced Auditory Serial Addition Test (Schwid et al., 2003). In two experiments, researchers found that both threat induced and state anxiety disrupted spatial but not verbal working memory on a three item n-back task in healthy undergraduates (Shackman et al., 2006). Given the fact that fatigue is the hallmark symptom in chronic fatigue syndrome, in addition to impairments in attention (Johnson, DeLuca, Diamond, & Natelson, 1998) and information processing (DeLuca et al., 2004), interventions that reduce fatigue and anxiety and improve vigilance could also potentially enhance information processing efficiency. MM is based on promoting a balance between a relaxed and vigilant state of mind (Wallace, 2006). The ability to self-regulate emotions has been found to be a key component in enhancing cognition (Austin, 1998; Moore & Malinowski, 2009). It is possible that the calming effects of MM combined with the increased capacity to focus on the present improved cognitive performance after brief training. MM training enhances present moment awareness by teaching participants to notice subtle distractions (feelings; thoughts; emotions) while repeatedly bringing attention back to the meditation object. This process can promote attentional stability (Epel, Daubenmier, Moskowitz, Folkman, & Blackburn, 2009; Wallace, 2006). Our findings provide robust evidence that brief MM training enhances sustained attention. The meditation group when compared to an active control group exhibited a greater number of extended hit runs on the two-back task. That is, the meditation group exhibited a significantly greater number of processing runs involving accurate and sustained working memory discriminations. The meditators were able to maintain focus and accurately retrieve information from working memory under conditions that require more rapid stimulus processing. Brief MM training improved vigilance and the efficiency of higher-order executive processes. Findings of improvements after MM training in visuo-spatial processing and in verbal fluency also indicated a greater efficiency in working and long-term memory retrieval in the meditators versus the control group. Some of these cognitive benefits have recently been reported with experienced meditators (Kozhevnikov et al., 2009). In fact, Brefczynski-Lewis and colleagues (2007) found that adept and novice meditators exhibited overlapping higher order attention-related neural activations. Similarly, 5 days of Integrative Body Mind Training effectively increased neural activity in the executive attention network which was correlated with better performance on attentional tasks (Tang et al., 2009) and twenty minutes of MM practice reduced habitual responding on the Stroop task (Wenk-Sormaz, 2005; Zeidan & Faust, 2008). Research associated with the benefits of brief MM training is sparse, but available evidence suggests that the immediate effects MM are not only associated with improving mood, but also developing deeper cognitive processing skills, specifically reducing lapses of attention. There were no differences between groups on digit span and these tasks were not timed, and did not, therefore, tax subjects on speed of processing. However, one factor that links the cognitive tasks that were affected by MM intervention is that they required sustained attention to perform well. Therefore, it appears that a short-term benefit of MM training could be increasing the ability to focus on timed or speeded tasks. Another explanation of why brief MM training improved cognition is associated with the ability to control the processing of self-referential thought. Some have provided evidence for overlapping networks between mindfulness, meta-awareness, executive functioning, and mind-wandering processes (Epel et al., 2009; Farb et al., 2007; Tang et al., 2007). Mind-wandering adversely affects cognitive performance by reducing supervision of goal-directed attention (Smallwood, McSpadden, & Schooler, 2007). When the mind wanders, discursive thoughts become the primary focus further decoupling attention directed towards the primary task (Smallwood & Schooler, 2006). This ‘‘automatic” process of mind-wandering suggests that there is a distinctive break in meta-awareness (Smallwood & Schooler, 2006). Meta-awareness is defined as the ability to reflect or be aware of ongoing thought or mental states (Epel et al., 2009; Smallwood et al., 2007). MM training is premised on teaching subjects to acknowledge discursive thoughts and to gently return their attention back to the meditation object. The immediate benefits of mindfulness meditation training may be associated with increasing the awareness of ongoing cognitive states, which improves attentional efficiency. As previously mentioned, the improvements in mood may have also improved information processing. Furthermore, recent findings suggest that improvements in mood may reduce mindwandering (Smallwood, Fitzgerald, Miles, & Phillips, 2009). In contrast, negative mood can lead to rumination and further lapses in attention. We postulate that the meditation group’s improvements in mood may have contributed to reducing mind-wandering, evidenced by significant improvements on a spectrum of cognitive tasks and mindfulness scores. Meta-awareness and executive functioning are independent but highly overlapping constructs (Fernandez-Duque et al., 2000). Although we did not directly test for this, the short-term benefits of mindfulness may promote reductions in reflexive, automatic processing of irrelevant information by improving attentional sustainability. Additionally, minimal training in mindfulness may be effective in promoting executive-level functioning in detecting when the mind has wandered (meta-
604
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
awareness), further reducing lapses in attention. Mindfulness practice promotes a form of meta-cognitive insight (Ortner, Kilner, & Zelazo, 2007), where MM practitioners learn to emotionally disengage from distracters (frustration; anxiety) (Teasdale, 1999). This form of top-down cognitive control leads the MM practitioner to more readily focus on the present task leading to better performance. Our findings extend previous work by using sensitive multiple functional and cognitive measures (e.g. the accuracy-selectable adaptive n-back task) that provide converging lines of evidence supporting the positive effect of MM on tasks that tap deep cognitive processing. Our findings apply to undergraduates and cannot be generalized to older adults, but show promise for positive effects on attentional tasks. Additionally, the improvements exhibited by the meditation group may be due to direct effects of just having meditated in session four. The Hobbit listening group was a more active control than prior control groups (e.g. sham mindfulness meditation) (Zeidan et al., in press), and may explain the lack of differences between groups on self-reported mood. However, the MM training group was significantly more effective at reducing anxiety and fatigue levels, when compared to the controls. We do not suggest that brief mental training is as effective as extensive training regimens. It is well documented that consistent and extensive meditation training promotes lasting changes in cognition and well-being (Cahn & Polich, 2006). Our findings show that there are immediate, short-term benefits to practicing mindfulness meditation. These benefits may have clinical implications. For instance, if a meditative state can be experienced after a brief training regimen, then individuals may feel more inclined to continue practice, which can lead to better health outcomes (Grossman et al., 2004). Moreover, meditation practice may be more attractive and easily disseminated if it can be shown to be effective without extensive training. This is the first study to demonstrate that 4 days of MM training can promote benefits on a range of cognitive tasks. Ongoing studies are employing neuroimaging to understand if and how brief meditation training affects the brain and behavior. Acknowledgments The authors would like to thank Susan Avett, Kari Young, Sasha Levons and Rachel Hinson for help with data collection and scoring. References Andreasen, A. K., Spliid, P. E., Andersen, H., & Jakobsen, J. (2009). Fatigue and processing speed are related in multiple sclerosis. European Journal of Neurology. doi:10.1111/j.1468-1331.2009.02776.x (Epub ahead of print). Austin, J. H. (1998). Zen and the brain: Toward an understanding of meditation and consciousness. Cambridge, MA: MIT Press. Baer, R., Smith, G., Hopkins, J., Krietemeyer, J., & Toney, L. (2007). Using self report assessment methods to explore facets of mindfulness. Assessment, 13, 27–45. doi:10.1177/1073191105283504. Benton, A. L. (1989). Multilingual aphasia examination. Iowa City, Iowa: AJA Associates. Brefczynski-Lewis, J. A., Lutz, A., Schaefer, H. S., Levinson, D. B., & Davidson, R. J. (2007). Neural correlates of attentional expertise in long-term meditation practitioners. Proceedings of the National Academy of Sciences, 104(27), 11483–11488. doi:10.1073/pnas.0606552104. Cahn, B. R., & Polich, J. (2006). Meditation states and traits: EEG, ERP, and neuroimaging studies. Psychological Bulletin, 132, 180–211. doi:10.1037/00332909.132.2.180. Carlson, L. E., Speca, M., Faris, P., & Patel, K. D. (2007). One year pre-post intervention follow-up of psychological, immune, endocrine and blood pressure outcomes of mindfulness-based stress reduction (MBSR) in breast and prostate cancer outpatients. Brain Behavior and Immunity, 21(8), 1038–1049. doi:10.1016/j.bbi.2007.04.002. Davidson, R. J., Kabat-Zinn, J., Schumacher, J., Rosenkranz, M., Muller, D., Santorelli, S. F., et al (2003). Alterations in brain and immune function produced by mindfulness meditation. Psychosomatic Medicine, 65, 564–570. doi:10.1097/01.PSY.0000077505.67574.E3. DeLuca, J., Christodoulou, C., Diamond, B. J., Rosenstein, E. D., Kramer, N., Ricker, J. H., et al (2004). The nature of memory impairment in chronic fatigue syndrome. Rehabilitation Psychology, 49(1), 62–70. Diamond, B. J., Johnson, S. K., Kaufman, M., & Graves, L. (2008). Relationships between information processing, depression, fatigue, and cognition in multiple sclerosis. Archives of Clinical Neuropsychology, 23, 189–199. doi:10.1016/j.acn.2007.10.002. Epel, E., Daubenmier, J., Moskowitz, J. T., Folkman, S., & Blackburn, E. (2009). Can meditation slow rate of cellular aging? Cognitive stress, mindfulness, and telomeres. Annals of the New York Academy of Sciences, 1172, 34–53. Farb, N. A., Segal, Z. V., Mayberg, H., Bean, J., McKeon, D., Fatima, Z., et al (2007). Attending to the present: Mindfulness meditation reveals distinct neural modes of self-reference. Social Cognitive Affective Neuroscience, 2(4), 313–322. doi:10.1093/scan/nsm030. Fernandez-Duque, D., Baird, J. A., & Posner, M. I. (2000). Executive attention and metacognitive regulation. Consciousness and Cognition, 9(2 Pt 1), 288–307. doi:10.1006/ccog.2000.0447. Grossman, P., Niemann, L., Schmidt, S., & Walach, H. (2004). Mindfulness-based stress reduction and health benefits: A meta-analysis. Journal Psychosomatic Research, 57, 35–43. doi:10.1016/S0022-3999(03)00573-7. Hölzel, B. K., Carmody, J., Evans, K. C., Hoge, E. A., Dusek, J. A., Morgan, L., et al (2009). Stress reduction correlates with structural changes in the amygdala. Social, Cognitive, and Affective Neuroscience. doi:10.1093/scan/nsp034 (Epub ahead of print). Jha, A. P., Krompinger, J., & Baime, M. J. (2007). Mindfulness meditation modifies subsystems of attention. Cognitive Affective Behavioral Neuroscience, 7(2), 109–119. doi:10.3758/CABN.7.2.109. Johnson, S. K., DeLuca, J., Diamond, B. J., & Natelson, B. (1998). Memory dysfunction in fatiguing illness: Examining interference and distraction in working memory. Cognitive Neuropsychiatry, 3(4), 269–285. Kozhevnikov, M., Louchakova, O., Josipovic, Z., & Motes, M. A. (2009). The enhancement of visuospatial processing efficiency through Buddhist deity meditation. Psychological Science, 20(5), 645–653. doi:10.1111/j.1467-9280.2009.02345.x. Lutz, A., Slageter, H. A., Rawlings, N. B., Francis, A. D., Greischer, L. L., & Davidson, R. J. (2009). Mental training enhances attentional stability: Neural and behavioral evidence. Journal of Neuroscience, 29(42), 13418–13427. doi:10.1523/JNEUROSCI.1614-09.2009. McNair, D., Loor, M., & Droppleman, L. (1971). Profile of mood states. San Diego, CA: Educational and Industrial Testing Service. Moore, A., & Malinowski, P. (2009). Meditation, mindfulness and cognitive flexibility. Consciousness and Cognition, 18(1), 176–186. doi:10.1016/ j.concog.2008.12.008. Nielsen, L., & Kaszniak, A. W. (2006). Awareness of subtle emotional feelings: A comparison of long-term meditators and nonmeditators. Emotion, 7(4), 392–405. doi:10.1037/1528-3542.6.3.392.
F. Zeidan et al. / Consciousness and Cognition 19 (2010) 597–605
605
Nyklícek, I., & Kuijpers, K. F. (2008). Effects of mindfulness-based stress reduction intervention on psychological well-being and quality of life: Is increased mindfulness indeed the mechanism? Annals of Behavioral Medicine, 35(3), 331–340. doi:10.1007/s12160-008-9030-2. Ortner, C. N., Kilner, S. J., & Zelazo, P. D. (2007). Mindfulness meditation and reduced emotional interference on a cognitive task. Motivation and Emotion, 31, 217–283. Posner, M. I., & Rothbart, M. K. (1998). Attention, self-regulation and consciousness. Philosophical Transactions of the Royal Society B: Biological Sciences, 353(1377), 1915–1927. doi:10.1098/rstb.1998.0344. Radloff, L. S. (1997). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385–401. doi:10.1177/014662167700100306. Schwid, S. R., Tyler, C. M., Scheid, E. A., Weinstein, A., Goodman, A. D., & McDermott, M. P. (2003). Cognitive fatigue during a test requiring sustained attention: A pilot study. Multiple Sclerosis, 9(5), 503–508. doi:10.1191/1352458503ms946oa. Shackman, A. J., Sarinopoulos, I., Maxwell, J. S., Pizzagalli, D. A., Lavric, A., & Davidson, R. J. (2006). Anxiety selectively disrupts visuospatial working memory. Emotion, 6(1), 40–61. doi:10.1037/1528-3542.6.1.40. Short, B. E., Kose, S., Mu, Q., Borckardt, J., Newberg, A., George, M. S., et al (2007). Regional brain activation during meditation shows time and practice effects: An exploratory FMRI Study. Evidence Based Complementary and Alternative Medicine: eCAM. doi:10.1093/ecam/nem163 (Epub ahead of print). Slagter, H. A., Lutz, A., Greischer, L. L., Nieuwenhuis, S., & Davidson, R. J. (2009). Theta phase synchrony and conscious target perception: Impact of intensive mental training. Journal of Cognitive Neuroscience, 8, 1536–1549. doi:10.1162/jocn.2009.21125. Smallwood, J., Fitzgerald, A., Miles, L. K., & Phillips, L. H. (2009). Shifting moods, wandering minds: negative moods lead the mind to wander. Emotion, 9(2), 271–276. 2009-04472-013 [pii]. Smallwood, J., McSpadden, M., & Schooler, J. W. (2007). The lights are on but no one’s home: Meta-awareness and the decoupling of attention when the mind wanders. Psychonomic Bulletin and Review, 14(3), 527–533. Smallwood, J., & Schooler, J. W. (2006). The restless mind. Psychological Bulletin, 132(6), 946–958. 2006-20202-006 [pii] 10.1037/0033-2909.132.6.946. Smith, A. (1982). Symbol-digit modalities test (SDMT) manual-revised. Los Angeles: Western psychological Services. Spielberger, C. D. (1983). Manual for the State-Trait Anxiety Inventory (STAI-form Y). Palo Alto, CA: Consulting Psychologists Press. Strauss, E., Sherman, E. M. A., & Spreen, O. (Eds.). (2006). A compendium of neuropsychological tests: Administration, norms, and commentary (pp. 546–677). New York: Oxford University Press. Tang, Y. Y., Ma, Y., Fan, Y., Feng, H., Wang, J., Feng, S., et al (2009). Central and autonomic nervous system interaction is altered by short-term meditation. Proceedings of the National Academy of Sciences, 106(22), 8865–8870. doi:10.1073/pnas.0904031106. Tang, Y. Y., Yinghua, W., Wang, J., Yaxin, F., Feng, S., Lu, Q., et al (2007). Short term meditation training improves attention and self-regulation. Proceedings of National Academy of Sciences, 104(43), 17152–17156. doi:10.1073/pnas.0707678104. Teasdale, J. D. (1999). Metacognition, mindfulness, and the modification of mood disorders. Clinical Psychology and Psychotherapy, 6, 146–155. doi:10.1002/ (SICI)1099-0879(199905)6:2<146::AID-CPP195>3.0.CO;2-E. Toneatto, T., & Nguyen, L. (2007). Does mindfulness meditation improve anxiety and mood symptoms? A review of the controlled research. Canadian Journal of Psychiatry, 52(4), 260–266. van Leeuwen, S., Muller, N. G., & Melloni, L. (2009). Age effects on attentional blink performance in meditation. Consciousness and Cognition 18(3), 593-599.
. Walach, T., Buchheld, N., Buttenmuller, V., Kleinknecht, N., & Schmidt, S. (2006). Measuring mindfulness: The Freiburg Mindfulness Inventory (FMI). Personality and Individual Differences, 40(8), 1543–1555. doi:10.1016/j.paid.2005.11.025. Wallace, A. (2006). The attention revolution: Unlocking the power of the focused mind. Boston, MA: Wisdom Publications. Wechsler, D. (1981). Weschler adult intelligence scale – Revised manual. New York: Psychological Corporation. Wenk-Sormaz, H. (2005). Meditation can reduce habitual responding. Alternative Therapies in Health Medicine, 11(2), 42–58. Zeidan, F., & Faust, M. (2008). The effects of brief mindful training on cognitive control. In Southeastern psychological association conference, Charlotte, NC. Zeidan, F., Johnson, S. K., Gordon, N. S., & Goolkasian, P. (in press). The effects of brief and sham mindfulness meditation on mood and cardiovascular variables. Journal of Alternative and Complementary Medicine. Zeidan, F., Gordon, N. S., & Goolkasian, P. (2009). The effects of brief meditation training on experimentally induced pain perception. Journal of Pain. doi:10.1016/j.jpain.2009.07.015 (Epub ahead of print).
Consciousness and Cognition 19 (2010) 606–616
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Aging and implicit memory: Examining the contribution of test awareness Lisa Geraci *, Terrence M. Barnhardt Department of Psychology, Texas A&M University, College Station, TX 77843-4235, United States
a r t i c l e
i n f o
Article history: Received 20 June 2009 Available online 18 April 2010 Keywords: Implicit memory Aging Priming Test awareness Post-test questionnaire Contamination
a b s t r a c t The study examined whether test awareness contributes to age effects in priming. Younger and older adults were given two priming tests (word-stem completion and category production). Awareness was assessed using both a standard post-test questionnaire and an on-line measure. Results from the on-line awareness condition showed that, relative to older adults, younger adults showed higher levels of priming and awareness, and a stronger relationship between the two, suggesting that awareness could account for age differences in priming. In contrast, in the post-test questionnaire condition, there was no age effect in word-stem completion or category production priming, despite the fact that awareness was greater in younger than older adults in the word-stem completion test and that category production priming was dependent on awareness in both age groups. These results suggest that awareness may mediate age effects in priming, but only under conditions of relatively high levels of awareness. Ó 2010 Elsevier Inc. All rights reserved.
0. Introduction There are clear effects of aging on memory performance when memory is measured explicitly by using tests that directly require people to recall the past, such as free recall tests (see Light (1996) for review). The picture is less clear when memory is tested implicitly or indirectly. Implicit memory tests require participants to perform a task that is predictably influenced by past experience although participants are presumably unaware of the past’s influence on their behavior. The resultant change in behavior (often improvements in accuracy or speed) is called priming (Graf & Schacter, 1985; Tulving, Schacter, & Stark, 1982). Sometimes priming is influenced by aging, but other times priming appears to be entirely unaffected by aging (see Fleischman and Gabrieli (1998), LaVoie and Light (1994), Light, Prull, La Voie, and Healy (2000) for reviews). Several test factors have been hypothesized to account for the mixed pattern of aging effects on implicit tests, yet no clear support has emerged for any of these explanations. For example, researchers have suggested that age effects occur on implicit tests that require participants to analyze the conceptual (or semantic) features of the stimulus, but not on tests that require participants to analyze the perceptual (e.g., physical word features) of the stimulus (Jelicic, 1995; Jelicic, Craik, & Moscovitch, 1996; Rybash, 1996). However, sometimes age effects do occur on perceptual tests, such as word-stem completion, in which participants are given the first few letters of a word at test and are required to complete the stem with the first word that comes to mind (Chiarello & Hoyer, 1988; Davis et al., 1990; Fleischman et al., 1999; Hultsch, Masson, & Small, 1991; Light & Singh, 1987; Winocur, Moscovitch, & Stuss, 1996). Conversely, sometimes age effects do not occur on conceptual tests such as category verification, in which participants are given categories and are asked to quickly indicate whether target items are members of the category (e.g., Light, Prull, & Kennison, 2000; Small, Hultsch, & Masson, 1995). With this in mind, researchers have instead suggested that age effects occur on tests that require participants to produce a response, but not on tests that require participants to simply identify the correct response (Gabrieli et al., 1994, 1999; Vaidya * Correspondence to: Department of Psychology, Texas A&M University, College Station, TX 77843-4235, United States. E-mail address: [email protected] (L. Geraci). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.015
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
607
et al., 1997). However, some research also shows that age effects do not occur on tests that require the production of a response when other factors are held constant (Geraci, 2006; Prull, 2004). Another interpretation is that age effects in priming occur on tests that entail response competition (i.e., where there are many possible correct responses) and not on those that entail little response competition (i.e., where there are very few, or sometimes only one, possible correct response; Gabrieli et al., 1999; Nyberg, Winocur, & Moscovitch, 1997; Vaidya et al., 1997; see also Light et al., 2000)). Yet, when responses competition is selectively manipulated, age effects do not always occur under conditions of high response competition (Geraci & Hamilton, 2009). Finally, several other factors in addition to test factors, including time of day (May, Hasher, & Foong, 2005) and individual differences in older adults’ neurological status (Fleischman & Gabrieli, 1998) have been offered as explanations for age effects in priming, but again there are results that appear inconsistent with these views (e.g., Geraci, 2006; Yang, Hasher, & Wilson, 2007). Yet another possibility is that age effects in priming depend on whether explicit memory processes inadvertently contribute to implicit memory performance (Habib, Jelicic, & Craik, 1996; Light, 1991; Mitchell, 1995; Mitchell & Bruss, 2003; Russo & Parkin, 1993). The idea is that a certain number of participants become aware of the connection between the studied and test items and then begin to use explicit or intentional memory processes to perform the ostensibly implicit task (see MacLeod (2008) for review). This phenomenon is sometimes referred to as ‘‘explicit contamination”. The contribution of explicit memory to implicit test performance is particularly problematic for understanding age effects in priming because younger and older adults differ in their ability to use explicit memory strategies (see Light (1996) for review). If younger adults are more likely to become test aware and/or engage in explicit memory strategies, then it is possible that age effects (when they are obtained) are driven by younger adults’ use of explicit memory strategies to augment performance on a supposedly implicit test. This possibility makes it difficult to interpret age differences in implicit test performance. Do they reflect true differences in priming or differences in explicit memory performance? Unfortunately, test awareness is not always assessed in age comparisons of priming. For test awareness to mediate age effects in priming, awareness would need to lead to more priming in general. In addition, either awareness would need to be greater in younger adults than it is in older adults or, when awareness occurs, it would have to increase priming more for younger adults than older adults. There is some evidence for the first idea that awareness increases priming. For example, when younger adults are given a post-test questionnaire to assess their awareness of the connection between studied and test items, younger adults who are classified as ‘‘test aware” on the basis of their responses to the questionnaire show more priming than ‘‘test unaware” participants (e.g., Barnhardt, 2004; Barnhardt & Geraci, 2008). In addition, some memory effects that are typically obtained on explicit tasks, such as free recall, are obtained for test aware participants, but not test unaware participants (e.g., Geraci & Rajaram, 2002). Together, these two types of evidence (i.e., showing that awareness leads to increases in priming and that awareness leads to differential priming effects) suggest that awareness can influence priming, at least in younger adults. There is less research examining awareness in older adults, but some studies show that older adults are less likely to report test awareness than younger adults. In a recent study examining the effects of aging and frontal lobe functioning on priming, over half of the younger adults reported test awareness on a post-test questionnaire, while only three (all with relatively high frontal functioning) of the 56 older adults reported test awareness (Geraci, 2006). Because there were so few test aware older adults, the effect of awareness on priming for this group could not be statistically examined. However, awareness did appear to change the level of priming for younger adults, although the difference was not significant. Other studies have also found relatively high levels of awareness in younger adults and low levels of awareness in older adults using posttest questionnaires (Geraci & Hamilton, 2009). Older adults may report less test awareness than younger adults because older adults actually experience less test awareness than younger adults or because older adults are less able to accurately recall their mental state than younger adults when the awareness questionnaire is administered sometime later. We return to this second possibility shortly. The final possibility—that awareness in younger adults leads to increases in priming whereas awareness in older adults does not—is difficult to test because so few older adults report awareness using the post-test questionnaire method of assessment. A handful of studies have compared age effects in priming by awareness using post-test questionnaires, but the data are mixed (e.g., Light & Albertson, 1989; Mitchell & Bruss, 2003; Park & Shaw, 1992). And again, lower reports of awareness from older adults may indicate less test awareness or poor awareness recall. One way to circumvent the potential problem of recalling one’s state of awareness from the past is to assess awareness at the time of testing, rather than waiting until some time later. In the current study, younger and older adults were given both an on-line awareness measure and a standard post-test questionnaire. For the on-line measure, participants were given standard implicit memory instructions to complete the task with the first word that comes to mind. They were also asked to note immediately after producing a word whether they thought the response they wrote might have been presented in the earlier study list (see Richardson-Klavehn and Gardiner (1996), for a similar paradigm used to assess involuntary awareness). If older adults’ relatively low frequency of awareness is not attributable to older adults forgetting their past mental state, then we would expect greater levels of awareness in younger than in older adults in the on-line condition. Further, using the on-line measure allowed us to examine age effects in priming under conditions in which participants are, by definition, test aware. Participants in the on-line condition can be considered test aware because the on-line test instructions essentially inform participants that the implicit memory test can be completed with previously studied words. Thus, the use of both the on-line assessment and the post-test questionnaire allowed us to contrast the pattern of age effects in priming under conditions of relatively high awareness (in the on-line condition) and relatively low awareness (in the post-
608
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
test questionnaire condition). It may be that age effects in priming occur only under circumstances in which there is a relatively high degree of awareness – as in the on-line condition – and that a relatively high degree of awareness is a kind of precondition for the contribution of explicit memory to implicit memory performance. We also used two different priming tasks, word-stem completion and category production. We used these two tasks because both tests are production tests (tests that require production of a single response where many possible responses are possible) and this type of test has been linked with age differences in priming (Gabrieli et al., 1999). We used two types of production tests: a word-stem completion task that is often classified as a perceptual test and a category production task that is classified as a conceptual test (see Roediger and McDermott (1993) for review). One might predict that awareness would be more likely to influence priming on conceptual tests, than on perceptual tests, because conceptual processes generally aid explicit memory performance (Craik & Lockhart, 1972).
1. Method 1.1. Design The experiment used a 2 2 2 mixed factorial design in which age (younger and older) served as the between-subjects factor and test type (word-stem completion and category production) and awareness measure (post-test questionnaire and online) served as the within-subjects variables. The order of the implicit memory tests was counterbalanced across participants, but the post-test questionnaire measure was always administered after the first test and the on-line probe was always administered during the second test. For example, when the stem completion test was first and the category production test was second, the stem completion test was followed by the awareness questionnaire and the category production test was accompanied by the on-line awareness procedure. Participants always received the post-test questionnaire condition first as an attempt to keep that implicit test as pure as possible. Because the participants in the on-line awareness condition were, by design, aware of a potential connection between study and test, it was assumed that having already received a post-test questionnaire for the previous test would have little influence on subsequent on-line performance. 1.2. Participants Sixty-four younger adults (M age = 19.19, SD = 1.14) and 64 older adults (M age = 74.25, SD = 5.74) participated in the experiment. The younger adults were recruited through the Psychology Department undergraduate participant pool and received credit toward the research participation component of their introductory psychology course. The older adults were mostly Texas A&M alumni recruited from the community. They received an honorarium of $10 in appreciation of their participation. Although 64 older adults were tested, it was later discovered that two of those participants had previously participated in the same experiment. These two participants were eliminated from the data analyses, leaving 64 younger adults and 62 older adults. Younger and older adults were given the Shipley Vocabulary Test (Zachary, 1986) to assess word knowledge. As expected, older adults had significantly higher vocabulary scores (M = 35.09, SD = 3.44) than younger adults (M = 31.09, SD = 3.92), F(1, 124) = 37.04, MSE = 13.60. Education level was also significantly higher for older adults (M = 16.48, SD = 2.53) relative to younger adults (M = 13.69, SD = .73), F(1, 124) = 72.15, MSE = 3.41. Older adults were also given a Mini-Mental State Exam (MMSE; Folstein, Folstein, & McHugh, 1975) to exclude from the analyses people with significant impairments in cognitive functioning. The average MMSE was 28.71 (SD = 1.11), and no one in the sample scored lower than 26 on this test. 1.3. Materials Two study lists of 70 items each were constructed such that in each of the lists there were 30 items that would latter serve as studied solutions on the word-stem completion test and 40 category exemplars (eight items in each of five categories) that could be used during the category production test. When one list was used for study, the items from the other list served as the baseline. This yielded 60 stems (30 studied and 30 baseline) for the word-stem completion test and 10 category labels (five studied and five baseline) for the category production test. Following guidelines for limiting potential test awareness (see MacLeod, 2008; Roediger & Geraci, 2005), 30 filler items were included in the word-stem completion test: eight were presented at the beginning of the test and the other 22 were randomly interspersed amongst the critical studied and baseline stems. In the end, the word-stem completion test consisted of 90 items. For the category production test, five filler category labels were included: two were presented at the beginning of the test and the other three were randomly interspersed amongst the studied and baseline category labels. In the end, the category production test consisted of 15 category labels. Finally, four buffer items were included at the beginning and end of each study list to blunt primacy and recency effects. All stems were unique in the stimulus set, including those from the buffers, fillers, and category production exemplars. Special care was taken to ensure that none of the word-stem completion stimuli could serve as exemplars in the category production test. In addition, based on their use in previous experiments, we selected critical items for both tests that would yield similar baseline performance across younger and older adult participants. To anticipate our results, baseline performance was similar across age groups. Please note, that all stimuli are available to readers upon request.
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
609
1.4. Procedure For the study task, participants were told that they would be presented with a list of words displayed one at a time on a computer screen and that their memory for these words would be tested later with a free recall test. After the study list was presented, participants were given a 5 min distracter task in which they were asked to mentally rotate objects to answer same-different judgments. Next, participants were given either the word-stem completion test or the category production test, depending on test order, and this test was disguised as yet another distracter task before the expected free recall test. For the word-stem completion test, participants were told that the test was designed to measure word knowledge. They were told that they would be given a list of word stems containing only the first three letters of words and that they should try to complete the stem with the first word that came to mind. Participants were given a booklet containing the word stems and they were told to work on completing one stem at a time. They were informed that speeded responding was of utmost importance and that if they could not complete an item within approximately five seconds, they should proceed to the next stem. Lastly, they were told not to use any proper nouns or words with fewer than five letters to complete the stems. In total, this test took approximately 7 min to complete. For the category production test, participants were told that this test was designed to measure their knowledge of categories. During the test, each category label (e.g., A Type of Bird) was presented on a separate page and participants were given 30 s to write down as many exemplars of the category as they could in that amount of time. At the end of the 30 s, participants heard a beep sounded by an audio cassette that prompted them to pull down the cover sheet and to proceed to the next category label. Participants were instructed to work on one test item at a time and not to go back or skip ahead in the booklet. This test took approximately 7 min. After the first test, whether it was stem completion or category production, participants were given a post-test questionnaire that was originally introduced by Bowers and Schacter (1990) and has been shown to accurately measure test awareness in younger adults (Barnhardt, 2004; Barnhardt & Geraci, 2008). When the test was stem completion, the questionnaire consisted of the following five questions: (1) What do you think was the purpose of the word-stem completion task that you just finished? (2) What was your general strategy in completing the word stems? (3) While you were doing the word-stem completion task, did you notice any relation between the words that were presented on the screen at the beginning of the experiment and the words you wrote in your booklet? (4) While you were doing the word-stem completion task, did you notice whether some of the words you wrote were the same as the words that had been displayed on the screen? (5) If you noticed that you were writing words that had been displayed on the screen, did you simply continue to use the first word that came to your mind or did you try to complete the stems with the words that had been displayed on the screen? The fifth question was designed to determine whether test aware participants continued to use the first word that came to mind or whether they changed their retrieval strategy to one in which they were intentionally trying to respond with studied words (see also Barnhardt, 2004; Mace, 2003, 2005). These questions were adjusted to fit with the category production test. The questions were printed on both sides of a single piece of paper, with the first three questions on one side and the last two questions on the other side and participants were instructed to answer the questions in order. If participants answered question five in a way that indicated they were intentionally retrieving the studied words, they were classified as ‘‘intentional” regardless of their other responses. If a participant was not classified as intentional, they were classified as aware if they indicated that they were aware on any of the first four questions. They were classified as unaware if they did not indicate awareness on any of the first four questions (see Barnhardt and Geraci (2008), for discussion of this classification). After participants completed the post-test questionnaire, they were given the second implicit test (either word-stem completion or category production) with on-line awareness instructions. If the second test was category production, participants were given the standard implicit test instructions described earlier and were also told: ‘‘As you are writing the category examples, if you notice any words that were ones you saw earlier on the computer screen, please circle these words as you go. Otherwise, just focus on writing as many category examples as you can in the allotted time.” Instructions were similar for the word-stem completion test. Finally, participants were given the expected free recall test in which they were told to write down all the words that they could remember from the study session. Younger adults were tested in groups of one to four and older adults were tested singly or in pairs. The entire procedure lasted approximately an hour and a half.
2. Results The results for the on-line and post-test measures are reported separately, starting with the on-line measure. For each measure, we examined age differences in awareness, followed by age differences in priming, and then the relationship between priming and awareness for that measure. Then we examined the relationship between age and awareness across the awareness measures. Finally, we present the free recall data. The significance level for all statistical tests was set at p < .05. 2.1. On-line awareness The number of participants who indicated awareness on at least one item in the on-line awareness condition is presented in Table 1. As can be seen, younger adults displayed more on-line awareness than older adults (91% of younger adults indi-
610
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
Table 1 Number of younger and older adults in each awareness category for both types of implicit tests in the on-line awareness condition. Younger
Zero OLA At least one OLA All participants
Older
WSC
CP
WSC
CP
6 26 32
0 32 32
16 14 30
18 14 32
Note: WSC = word-stem completion; CP = category production. OLA = on-line awareness. Standard deviations are in parentheses.
cated awareness on at least one item, whereas only 45% of older adults did so). Over half of the older adults failed to indicate awareness on any of the items in either the word-stem completion test or the category completion test. In contrast, only 19% of young adults did not indicate awareness on at least one stem completion response and all of the young adults indicated awareness on at least one category production response. Both of these age differences were significant, v2(1) = 8.09 and v2(1) = 25.04, respectively. Next we examined the false alarm rates for the test aware participants in the two age groups. In the stem completion test, the mean false alarm rate for younger adults was .27 and for older adults was .14. In the category production test, this rate was .03 for the younger adults and .07 for the older adults. Neither difference approached significance, p’s > .4. 2.2. Priming in the on-line awareness condition Next we examined whether there was an age effect in priming in the on-line awareness condition. Priming was calculated by subtracting the proportion of correct responses to test items with nonstudied solutions from the proportion of correct responses to test items with studied solutions. We first examined priming without regard to awareness (see all participants section of Table 2). Priming was significantly different from zero in all cells except for the stem completion test for older adults. Younger adults had significantly greater priming (M = .12) than older adults (M = .05), t(62) = 2.96, SE = .02 on the category production test. Younger adults also showed numerically more priming (M = .09) than older adults (M = .04) on the word-stem completion test, but this difference did not reach significance, p = .13. Collapsing across test type (category production and word-stem completion), there was a significant age effect in priming in the on-line awareness condition, (younger = .10, older = .04, t(124) = 2.98, SE = .02). 2.3. The relationship between awareness and priming in the on-line awareness condition To examine the relationship between awareness and priming, we calculated the correlation between the magnitude of priming and the proportion of studied responses on which awareness was indicated. Collapsing across test type, results showed that the correlation between priming and on-line awareness was significant in the younger adults (r = .43, n = 63), but not in the older adults (r = .19, n = 62, p = .34). The difference between these two correlations approached significance, Fisher’s z = 1.85, two-tailed p = .06. We have shown that a greater number of younger participants were aware, that younger participants displayed more priming, and that younger adults displayed a stronger positive relationship between the amount of awareness and the magnitude of priming. From this, we might expect that the percentage of studied responses circled as aware by younger adults would be greater than that for older adults. Interestingly, when the number of aware items was examined as a proportion of the number of studied responses in those participants that had circled at least one studied response, younger and older adults showed similar levels of awareness. In word-stem completion, younger adults were aware for 41% of items, older
Table 2 Priming for younger and older adults in each awareness category for both types of implicit tests in the on-line awareness condition. Younger
Older
WSC
CP
WSC
CP
Zero OLA Priming Baseline
.03 (.19) .16 (.11)
NA NA
.04 (.15) .25 (.10)
.04 (.08) .14 (.07)
At least one OLA Priming Baseline
.10 (.10) .21 (.08)
.12 (.10) .19 (.06)
.03 (.13) .25 (.10)
.07 (.09) .18 (.08)
All participants Priming Baseline
.09 (.12) .20 (.09)
.12 (.10) .19 (.06)
.04 (.14) .25 (.10)
.05 (.08) .16 (.07)
Note: WSC = word-stem completion; CP = category production. OLA = on-line awareness. Standard deviations are in parentheses.
611
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
adults were aware for 37% of items, and these means did not differ, t < 1. In category production, younger adults were aware for 65% of items, older adults were aware for 60% of items, and these means also did not differ, t < 1. Thus, it appears that, once the older adults recognized one item as having been studied, their level of awareness was very similar to that in younger adults. In sum, the on-line awareness data show that, relative to older adults, younger adults showed higher levels of awareness, higher levels of priming, and a stronger relationship between the magnitude of priming and the magnitude of on-line awareness. Thus, it appears that awareness, as measured using this on-line method, could account for age differences in priming. 2.4. Post-test questionnaire awareness First, we examined whether there were age or test differences in awareness using the post-test questionnaire. The number of participants in each awareness category, as a function of age and type of test, is displayed in the top half of Table 3. The most prominent feature of these data is that there was a difference between the stem completion and category production tests in the distribution of participants across the awareness categories. Participants were much less likely to be classified as intentionally retrieving following the stem completion test than after the category production test: only five participants were classified as intentional after stem completion, whereas 30 were so classified after category production. In complementary fashion, participants were much more likely to be classified as unaware following the stem completion test than after the category production test: 27 participants reported that they were unaware after stem completion, and only two reported that they were unaware after category production. Approximately the same number of participants reported that they were test aware after the two tests (32 vs. 30). A hierarchical log linear analysis of the three-way dependence of age, test type, and awareness classification verified a significant two-way dependence of test type and awareness, difference G2 = 45.49. This analysis confirmed the fact that, on the word-stem completion test, participants were largely either test unaware or test aware, whereas on the category production test, participants were either test aware or intentionally retrieving. Almost no one was classified as test unaware on the category production test. The other prominent feature of these data is that it appeared that younger adults reported more test awareness than older adults. However this effect did not reach significance, G2 = 5.065, p = .079. Given the scaling differences in awareness type across the two tests (word-stem completion and category production), we attempted to ‘‘standardize” and simplify awareness classification across the two test conditions in order to better compare level of awareness across the two age populations. This approach consisted of defining the ‘‘median” amount of awareness in the perceptual test as the break between unaware and aware (collapsing across age, 27 participants had been classified as unaware and 32 as aware). In the conceptual test, the median amount of awareness was defined as the break between aware and intentional (collapsing across age, 30 participants had been classified as aware and 30 as intentional). Those below the median were classified as ‘‘less aware” and those above the median were classified as ‘‘more aware”. The reclassification is presented at the bottom of Table 3. To instate this new classification in the word-stem completion condition, the awareness and intentional categories were collapsed into the ‘‘more aware” category. This resulted in five participants – 8% – originally classified as intentional being reclassified as ‘‘more aware”. In the category production condition, the unaware and aware categories were collapsed into the ‘‘less aware category”. This resulted in two participants – 3% – originally classified as unaware being reclassified as ‘‘less aware”. This recategorization eliminated the problem that some of the cells in the full design had a very small number of participants (e.g., only one young participant and one old participant were classified as unaware in the category production condition). Using this standardization and reclassification procedure, the clear pattern that emerges in Table 3 is that younger adults reported more awareness than older adults on the post-test questionnaire, thus replicating previous findings (e.g., Geraci, 2006). In word-stem completion, more of the younger adults were classified as more aware (23), while fewer of the younger adults were classified as less aware (9). However, the reverse was true for older adults (14 were classified as more aware and 18 as less aware). The pattern was similar, but much weaker, in the category production test condition: for the younger adults, 17 were classified as more aware and 15 as less aware, whereas 13 older adults were classified as more aware Table 3 Number of participants in each awareness category as function of age and type of test for the post-test questionnaire condition. Younger
Older
WSC
CP
WSC
CP
Awareness category Unaware Aware Intentional All participants
9 20 3 32
1 14 17 32
18 12 2 32
1 16 13 30
Reclassification Less aware More aware All participants
9 23 32
15 17 32
18 14 32
17 13 30
Note: WSC = word-stem completion; CP = category production. Standard deviations are in parentheses.
612
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
and 17 as less aware. A hierarchical log linear analysis of the three-way dependence of age, test type, and awareness classification yielded a significant two-way dependence of age and awareness, difference G2 = 4.57, p = .03. Collapsing across type of test, 62.5% of younger adults were classified as more aware (37.5% as less aware), whereas only 43.5% of older adults were classified as more aware (56.5% as less aware). These results demonstrate that younger adults were more test aware than older adults, as measured by the post-test questionnaire. Although the three-way interaction was not statistically significant, the fact that younger adults were more test aware than older adults appeared to be especially true in the stem completion condition. 2.5. Post-test questionnaire priming We also examined whether there was an age effect in priming in the post-test questionnaire condition. Results showed that there was no significant age effect in priming in the word-stem completion test (younger = .07, older = .04), t(62) = 1.23, SE = .03) or category production test (younger and older = .07), t(62) < 1). In addition, collapsing across test type showed no significant age difference in priming (younger = .07, older = .05), t(124) < 1. p = .39. Thus, we found a significant age effect in priming in the on-line awareness condition but not in the post-test questionnaire condition, despite the fact that there were age differences in awareness in both conditions. We examined possible baseline performance differences between the young and older adults because such differences can sometimes undermine a straight-forward interpretation of age effects in priming. For example, on the word-stem completion test, older adults (M = .24) had significantly higher baselines than younger adults (M = .19), t(124) = 3.07, SE = .02, which could indicate that reduced priming for older adults was an artifact of older adults having higher baseline performance. However, this explanation could not account for the data from the category production test, where the younger adults (M = .20) had a significantly higher baseline performance than the older adults (M = .16), t(124) = 3.01, SE = .01, yet priming was equivalent across the two age groups. Thus, baseline differences between the younger and older adults did not appear to account for the priming results. 2.6. The relationship between awareness and priming in the post-test questionnaire condition Next, we examined whether reports of awareness on the post-test questionnaire were related to priming levels for younger and older adults (see Table 4). Inspection of Table 4 shows that there was a clear relationship between priming and awareness classification for both younger and older adults in category production: priming was .10 for more aware younger adults, whereas it was .03 for less aware younger adults; priming was .10 for more aware older adults, whereas it was .04 for less aware older adults. A 2 2 (age aware) ANOVA showed that there was a main effect of awareness on category production priming, F(1, 58) = 7.07, MSE = .58. This finding contrasted with the lack of an age effect in awareness and priming in category production, a point we return to in the general discussion. Turning to stem completion priming, inspection of Table 4 shows that there was no relationship between priming and awareness classification for either the younger or the older adults: priming was .07 for both the less and more aware younger adults and was .04 for both the less and more aware older adults. Again, the fact that there was no relationship between stem completion priming and awareness contrasted with the fact that younger adults displayed a numerically greater amount of priming (although the advantage was not significant) and younger adults reported more awareness in the post-test questionnaire. Again, we return to this point in the general discussion. 2.7. Comparing on-line and post-test questionnaire awareness To determine whether older adults experienced less awareness than younger adults at the time of test or whether they forgot their state of awareness while waiting for the questionnaire to be administered, we examined the number of younger Table 4 Priming for younger and older adults as a function of level of awareness for both types of implicit test in the post-test questionnaire condition. Younger
Older
WSC
CP
WSC
CP
Less aware Priming Baseline
.07 (.09) .13 (.06)
.03 (.11) .20 (.09)
.04 (.11) .23 (.10)
.04 (.11) .16 (.08)
More aware Priming Baseline
.07 (.08) .20 (.08)
.10 (.11) .20 (.06)
.04 (.13) .23 (.13)
.10 (.07) .16 (.07)
All participants Priming Baseline
.07 (.08) .18 (.08)
.07 (.11) .20 (.07)
.04 (.12) .23 (.11)
.07 (.10) .16 (.07)
Note: WSC = word-stem completion; CP = category production. Standard deviations are in parentheses.
613
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
and older adults reporting awareness on these two measures. Table 5 shows the awareness counts as a function of awareness measure, implicit test, age, and awareness classification. For purposes of this comparison, aware and intentional categories in the questionnaire condition were collapsed into a single aware category to parallel the nature of the classification in the online condition. Visual inspection of Table 5 revealed quite different patterns for the stem completion and category production conditions. In the stem completion condition, as noted in the analyses above, younger adults were more aware than older adults. Most important, the nature of this age difference was remarkably similar across the on-line and questionnaire conditions, which is inconsistent with the notion that older adults report less awareness on standard post-test questionnaires than younger adults because they cannot recall their conscious states from the time of testing. Instead, this pattern of data indicated that older adults reported less awareness on standard post-test questionnaire than younger adults because they experienced less awareness at the time of testing. In the category production condition, there was once again a remarkable similarity in the amount of awareness reported by younger adults in the on-line and questionnaire conditions. However, this was not the case for older adults. Older adults reported more awareness in the post-test questionnaire condition than in the on-line condition. Thus, it appeared that older adults were willing to describe themselves as aware in a post-test questionnaire even though they were relatively unable to accurately identify – in the on-line condition – the studied items that they had produced. Possible reasons for this finding are discussed in the General Discussion section. Regardless of the exact reason for this pattern of data in the category production condition, overall, the results indicated that older adults were not under-reporting the extent of their test awareness on posttest questionnaires because they were forgetting that they had been aware at the time of the implicit test. 2.8. Free recall Lastly, we examined age effects in explicit memory. Results showed that there was an age effect on recall of both wordstem completion and category production items. Younger adults recalled more word stem items (M = .10, SD = .08) than older adults (M = .05, SD = .05), F(1, 124) = 24.03, MSE = .01, and they recalled more category production items (M = .22, SD = .12) than older adults (M = .15, SD = .13), F(1, 124) = 24.03, MSE = 10.59. Thus, we obtained the expected age effect in explicit memory. 3. General discussion This study examined the influence of test awareness on age effects in priming. We used two methods for assessing test awareness: an on-line assessment and a post-test questionnaire. Results showed that in the on-line assessment condition, overall, younger adults showed greater priming than older adults, greater awareness than older adults, and a stronger relationship between awareness and priming than older adults. These results suggested that greater priming in younger adults is associated with greater test awareness in younger adults. In contrast, results from the post-test questionnaire condition showed equivalent priming for younger and older adults in both the category production and stem completion tests. Awareness likely led to increased priming for younger adults in the on-line condition, but not the post-test questionnaire condition, because participants in the on-line condition were test-informed. That is, in the on-line awareness condition, participants were told that there could be some items from the study session on the test and to circle those items if they recognized any of their responses as studied. In addition, the on-line awareness condition always followed the post-test questionnaire and so participants already had had some experience with the nature of implicit tests, which may have facilitated their ability to identify studied responses. Further, the use of intentional study instructions may have facilitated better explicit memory in the younger adults. Previous research has shown that younger adults display better recall and recognition memory performance after intentional study compared to incidental study, while older adults do not show any benefit of intentional study (e.g., Mitchell & Perlmutter, 1986). Thus, the fact that we obtained an age effect in priming in the on-line awareness condition, but not in the post-test questionnaire condition, may be due to the fact that awareness levels were relatively high in the on-line awareness condition. Further, it appears that younger adults in this condition took advantage of their test awareness in a way that older adults did not, possibly by using explicit retrieval strategies to boost their performance. Indeed, it appeared that the relatively small
Table 5 Number of younger and older adults in each awareness category for both types of implicit tests in both awareness assessment conditions. Online
Questionnaire
WSC
Unaware Aware
CP
WSC
CP
Young
Old
Young
Old
Young
Old
Young
Old
6 26
16 14
0 32
18 14
9 23
18 14
1 31
1 29
Note: In the post-test questionnaire condition, aware and intentionally retrieving participants were collapsed into a single aware category.
614
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
number of older adults that identified at least one response as having been studied may have been attributable to the relative inability of the older adults to initiate some kind of recollective processing during the on-line task. In contrast, in the post-test questionnaire condition, the opportunity for awareness was much lower, particularly given all of the additional standard procedures we used to limit awareness, including using a cover story and using additional nonstudied, ‘‘filler” items at test. With these procedures in place, we hoped to mimic the standard ‘‘best case scenario” in which the influences of test awareness would presumably be relatively low. To be clear, when awareness categories were compared across the questionnaire and on-line conditions (see Table 5), the number of participants in each cell was very similar, except for older adults in the category production condition, where there were actually more aware older adults in the questionnaire condition than in the on-line condition. We believe this is an artifact of the assessment methods themselves. In order to be categorized as ‘‘aware” in the on-line condition, a participant must accurately identify at least one instance of having responded with a studied word, a relative strict criterion for being classified as aware. In contrast, in the questionnaire method, a participant can be described as aware without having to identify a specific instance of having said a studied word, a relatively liberal criterion for being classified as aware. In many instances, participants in the questionnaire condition were unable to specify an instance of awareness when asked. In the present experiment, the last question on the awareness questionnaire asked participants to identify one of the responses they had made that they thought was a studied word. For younger adults, only five of the 23 aware participants in the stem completion test and only 16 of the 31 aware participants in the category production test were able to do so For older adults, only two of the 14 aware participants in the stem completion test and only 15 of 29 in the category production test could report an aware item. Thus, despite the relatively equivalent numbers of aware participants across the questionnaire and on-line conditions for both the younger and older age groups, it still seems fair to say that awareness, in general, was greater in the on-line condition than in the questionnaire condition, and that the greater incidence/degree of awareness in the on-line condition set the stage for meeting the preconditions necessary for observing age differences in priming. Taken together then, the on-line and post-test questionnaire results indicated that test awareness may have been a factor leading to or associated with age effects in priming when the potential of awareness was relatively high, as in the on-line awareness condition. When test awareness was relatively low, as in the post-test questionnaire condition, priming differences across younger and older adults did not appear. Despite the fact that there was no age difference in priming in the post-test questionnaire condition, both younger and older adults did report some test awareness and, at least in the stem completion test, this awareness was greater in younger adults. This finding is noteworthy for two reasons. First, this pattern of data (showing greater awareness in younger adults relative to older adults) was similar to the pattern obtained in the on-line awareness condition. Given the same age-associated pattern of awareness across the two tests, this finding suggests that post-test questionnaires are probably fairly accurate measures of the degree of awareness that was present during testing. Moreover, this finding is important for understanding age effects in awareness, because it suggests that the lower reports of awareness from older adults, relative to younger adults, is not due to problems with retrospective accounts of awareness that could be exacerbated for older adults (such as increased forgetting of awareness states). Instead, our results show that older adults are also less aware at the time of testing than are younger adults. Second, the finding that awareness levels differed for younger and older adults in the post-test questionnaire data, at least in the stem completion condition, is interesting in light of the fact that this awareness did not lead to or result from an age difference in priming. We have suggested that this pattern of data occurred because the overall level of awareness was relatively low in this condition. In neither the stem completion nor the category production tests were both the hypothesized preconditions met. In the stem completion condition, younger adults were more aware than older adults (one of the preconditions), but increases in awareness were not associated with increases in priming (the other precondition). In the category production test, the problem was just the reverse: increases in awareness were associated with increases in category production priming (one of the postulated preconditions), but the younger adults were not more aware than the older adults (the other postulated precondition). The absence of an awareness advantage for younger adults in the category production condition may have been due to a ceiling effect (i.e., nearly 100% of older adults were classified as aware; as a result, it was difficult for more younger adults to be aware). Again, it was only in the on-line condition that both preconditions were met – greater awareness in younger adults than older adults and an association between awareness and priming (with said association also greater in younger adults) – and age effects in priming were observed. With regard to the post-test questionnaire condition, the association between awareness and priming was greater in category production than in stem completion. This finding was not surprising given the different processing demands of the two tests. The category production test is typically classified as a conceptual test, meaning that priming on this test is influenced by processing the meaning of the studied items, while the word-stem completion test is typically classified as a perceptual test, meaning that priming on this test is influenced by processing the perceptual or surface features of the studied items (see Roediger, 1990). Previous research shows a much stronger relationship between test awareness and conceptual, or deep, processing than between test awareness and perceptual, or shallow processing (e.g., Barnhardt & Geraci, 2008; Graf, Mandler, & Haden, 1982; Mace, 2003, 2005; Richardson-Klavehn, Gardiner, & Java, 1994; Toth, Reingold, & Jacoby, 1994). Thus, it appears that awareness is probably more highly associated with priming under conditions of conceptual processing, than under conditions of perceptual processing. Certainly we have very strong evidence that under conditions where younger adults show more priming than older adults (the on-line awareness condition in the current study), younger adults are also more test aware than older adults. This finding provides support for the idea that awareness may mediate age effects in priming when they are obtained. Most research-
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
615
ers assume that awareness leads to increased priming via the following processes. Participants notice the study-test connection, which should be more likely for younger adults who have better explicit memory than older adults. Once participants become aware of the fact that some test items can be completed with studied items they, in turn, begin to treat the implicit test as an explicit one. Because younger adults are also better able to engage in explicit strategies to boost performance than older adults, they are better able to complete the implicit task with studied items than are older adults. However, an alternate possibility is that awareness is associated with priming because younger adults are more likely to use test items to complete stems or category cues, and so they have more opportunities to become test aware. By this view, part of the phenomenon of priming might simply include some level of awareness (which could be conceived of as involuntary aware memory; Kinoshita, 2001; Richardson-Klavehn & Gardiner, 1995, 1996). This does not mean that people attempt to recall the past, but simply that good memory performance is associated with more explicit memory features. Finally, the idea that explicit memory contributes to age effects in priming can accommodate some specific patterns in the literature. For example, one possible explanation for the previously reported dissociation between older adults’ performance on production and identification implicit tests could be that younger adults may be more likely to use explicit strategies to perform production tests than identification tests (see Geraci (2006) for this hypothesis). Production tests are often less speeded than identification tests, which may allow for a greater opportunity for awareness to influence priming. The influence of awareness on this and other test dissociations awaits future testing. For now, though, the current data suggest that under some conditions, awareness may contribute to age effects in priming, which may help explain why age effects in priming are only sometimes obtained. Acknowledgment We would like to thank Elizabeth Gomez for her help with data collection. References Barnhardt, T. M. (2004). Different involuntary mechanisms underlie priming and LOP effects in stem completion tests. Memory, 12, 614–636. Barnhardt, T. M., & Geraci, L. (2008). Are post-test questionnaires valid? Investigating the use of posttest questionnaires for assessing awareness in implicit memory tests. Memory & Cognition, 36, 53–64. Bowers, J. S., & Schacter, D. L. (1990). Implicit memory and test awareness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 404–416. Chiarello, C., & Hoyer, W. J. (1988). Adult age differences in implicit and explicit memory: Time course and encoding effects. Psychology and Aging, 3, 358–366. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684. Davis, H. P., Cohen, A., Gandey, M., Colombo, P., Van Dusseldorp, G. V., Simolke, N., et al (1990). Lexical priming deficits of the function of age. Behavioral Neuroscience, 104, 288–297. Fleischman, D. A., & Gabrieli, J. D. E. (1998). Repetition priming in normal aging and Alzheimer’s disease: A review of findings and theories. Psychology and Aging, 13, 88–119. Fleischman, D. A., Gabrieli, J. D. E., Gilley, D. W., Hauser, J. D., Lange, K. L., Dwornik, L. M., et al (1999). Word-stem completion priming in healthy aging and Alzheimer’s disease: The effects of age, cognitive status, and encoding. Neuropsychology, 13, 22–30. Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). Mini-Mental State: A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12, 189–198. Gabrieli, J. D. E., Keane, M. M., Stanger, B. Z., Kjelgaard, M. M., Corkin, S., & Growdon, J. H. (1994). Dissociations amongst structural-perceptual, lexical semantic, and event-fact memory systems in Alzheimer, amnesic, and normal subjects. Cortex, 30, 75–103. Gabrieli, J. D. E., Vaidya, C. J., Stone, M., Francis, W. S., Thompson-Schill, S. L., Fleischman, D. A., et al (1999). Convergent behavioral and neuropsychological evidence for a distinction between identification and production forms of repetition priming. Journal of Experimental Psychology: General, 128, 479–498. Geraci, L. (2006). A test of the frontal lobe functioning hypothesis of age deficits in production priming. Neuropsychology, 20, 530–548. Geraci, L., & Hamilton, M. (2009). Examining the response competition hypothesis of age effects in implicit memory. Aging, Neuropsychology, and Cognition, 16, 683–707. Geraci, L., & Rajaram, S. (2002). The orthographic distinctiveness effect on direct and indirect tests of memory: Delineating the awareness and processing requirements. Journal of Memory and Language, 47, 273–291. Graf, P., Mandler, G., & Haden, P. E. (1982). Simulating amnesic symptoms in normal subjects. Science, 218, 1243–1244. Graf, P., & Schacter, D. L. (1985). Implicit and explicit memory for new associations in normal and amnesic subjects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 501–518. Habib, R., Jelicic, M., & Craik, F. I. M. (1996). Are implicit memory deficits in the elderly due to differences in explicit memory performance? Aging, Neuropsychology, & Cognition, 3, 264–271. Hultsch, D. F., Masson, M. E. J., & Small, B. J. (1991). Adult age differences in direct and indirect tests of memory. Journal of Gerontology: Psychological Sciences, 46, 22–30. Jelicic, M. (1995). Aging and performance on implicit memory tasks: A brief review. International Journal of Neuroscience, 82, 155–161. Jelicic, M., Craik, F. I. M., & Moscovitch, M. (1996). Effects of aging on different explicit and implicit memory tasks. European Journal of Cognitive Psychology, 8, 225–234. Kinoshita, S. (2001). The role of involuntary aware memory in the implicit stem completion and fragment completion tasks: A selective review. Psychonomic Bulletin & Review, 8, 58–69. LaVoie, D., & Light, L. L. (1994). Adult age differences in repetition priming: A meta-analysis. Psychology and Aging, 9, 539–553. Light, L. L. (1991). Memory and aging: Four hypotheses in search of data. Annual Review of Psychology, 42, 333–376. Light, L. L. (1996). Memory and aging. In E. J. Bjork & R. A. Bjork (Eds.), Memory (pp. 443–490). San Diego: Academic Press. Light, L. L., & Albertson, S. A. (1989). Direct and indirect tests of memory for category exemplars in young and older adults. Psychology and Aging, 4, 487–492. Light, L. L., Prull, M. W., & Kennison, R. F. (2000). Divided attention, aging, and priming in exemplar generation and category verification. Memory & Cognition, 28, 856–872. Light, L. L., Prull, M. W., La Voie, D., & Healy, M. R. (2000). Dual-process theories of memory in older age. In T. J. Perfect & E. A. Maylor (Eds.), Models of cognitive aging (pp. 238–300). Oxford, England: Oxford University Press. Light, L. L., & Singh, A. (1987). Implicit and explicit memory in young and older adults. Journal of Experimental Psychology: Learning, Memory and Cognition, 13, 531–541. Mace, J. H. (2003). Involuntary aware memory enhances priming on a conceptual implicit memory task. American Journal of Psychology, 116, 281–290.
616
L. Geraci, T.M. Barnhardt / Consciousness and Cognition 19 (2010) 606–616
Mace, J. H. (2005). Experimentally manipulating the effects of involuntary conscious memory on a priming task. American Journal of Psychology, 118, 159–182. MacLeod, C. M. (2008). Implicit memory tests: Techniques for reducing conscious intrusion. In J. Dunlosky & R. A. Bjork (Eds.), Handbook of metamemory and memory: Papers in honor of Thomas O. Nelson (pp. 245–263). New York: Psychology Press. May, C. P., Hasher, L., & Foong, N. (2005). Implicit memory, age, and time of day: Paradoxical priming effects. Psychological Science, 16, 96–100. Mitchell, D. B. (1995). Semantic processes in implicit memory: Aging with meaning. In T. R. Bashore & P. A. Allen (Eds.), Age differences in word and language processes (pp. 110–142). Amsterdam, Netherlands: North-Holland Elsevier. Mitchell, D. B., & Bruss, P. J. (2003). Age differences in implicit memory: Conceptual, perceptual, or methodological? Psychology and Aging, 18, 807–822. Mitchell, D. B., & Perlmutter, M. (1986). Semantic activation and episodic memory: Age similarities and differences. Developmental Psychology, 22, 85–94. Nyberg, L., Winocur, G., & Moscovitch, M. (1997). Correlation between frontal lobe functions and explicit and implicit stem completion in healthy elderly. Neuropsychology, 11, 70–76. Park, D. C., & Shaw, R. J. (1992). Effect of environmental support on implicit and explicit memory in younger and older adults. Psychology and Aging, 7, 632–642. Prull, M. W. (2004). Exploring the identification-production hypothesis of repetition priming in young and older adults. Psychology and Aging, 19, 108–124. Richardson-Klavehn, A., & Gardiner, J. M. (1995). Retrieval volition and memorial awareness in stem completion: An empirical analysis. Psychological Research, 57, 166–178. Richardson-Klavehn, A., & Gardiner, J. M. (1996). Cross-modality priming in stem completion reflects conscious memory, but not voluntary memory. Psychonomic Bulletin & Review, 3, 238–244. Richardson-Klavehn, A., Gardiner, J. M., & Java, R. I. (1994). Involuntary conscious memory and the method of opposition. Memory, 2, 1–29. Roediger, H. L. (1990). Implicit memory: Retention without remembering. American Psychologist, 45, 1043–1056. Roediger, H. L., & Geraci, L. (2005). Implicit memory tasks in cognitive research. In A. Wenzel & D. Rubin (Eds.), Cognitive methods and their application to clinical research (pp. 129–151). Washington, DC: APA Books. Roediger, H. L., & McDermott, K. B. (1993). Implicit memory in normal human participants. In F. Boller & J. Grafman (Eds.). Handbook of neuropsychology (Vol. 8, pp. 63–131). Amsterdam: Elsevier. Russo, R., & Parkin, A. J. (1993). Age differences in implicit memory: More apparent than real. Memory & Cognition, 21, 73–80. Rybash, J. M. (1996). Implicit memory and aging: A cognitive neuropsychological perspective. Developmental Neuropsychology, 12, 127–179. Small, B. J., Hultsch, D. F., & Masson, M. E. J. (1995). Adult age differences in perceptually based, but not conceptually based implicit tests of memory. Journal of Gerontology, 50, 162–170. Toth, J. P., Reingold, E. M., & Jacoby, L. L. (1994). Toward a redefinition of implicit memory: Process dissociations following elaborative processing and selfgeneration. Journal of Experimental Psychology: Learning, Memory, & Cognition, 20, 290–303. Tulving, E., Schacter, D. L., & Stark, H. A. (1982). Priming effects in word-fragment completion are independent of recognition memory. Journal of experimental Psychology: Learning, Memory, and Cognition, 8, 336–442. Vaidya, C. J., Gabrieli, J. D. E., Keane, M. M., Monti, L. A., Gutierrez-Rivas, H., & Zarella, M. M. (1997). Evidence for multiple mechanisms of conceptual priming on implicit memory tests. Journal of Experimental Psychology: Learning, Memory and Cognition, 23, 1324–1343. Winocur, G., Moscovitch, M., & Stuss, D. T. (1996). Explicit and implicit memory in the elderly: Evidence for double dissociation involving medial-temporaland frontal-lobe functions. Neuropsychology, 10, 57–65. Yang, L., Hasher, L., & Wilson, D. E. (2007). Synchrony effects in automatic and controlled retrieval. Psychonomic Bulletin & Review, 14, 51–56. Zachary, R. A. (1986). Shipley Institute of Living Scale, revised manual. Los Angeles, CA: Western Psychological Services.
Consciousness and Cognition 19 (2010) 617–626
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Cues for self-recognition in point-light displays of actions performed in synchrony with music Vassilis Sevdalis *, Peter E. Keller Music Cognition and Action Group, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
a r t i c l e
i n f o
Article history: Received 7 July 2009 Available online 9 April 2010 Keywords: Action perception Movement kinematics Self-recognition Audiovisual perception Music
a b s t r a c t Self-other discrimination was investigated with point-light displays in which actions were presented with or without additional auditory information. Participants first executed different actions (dancing, walking and clapping) in time with music. In two subsequent experiments, they watched point-light displays of their own or another participant’s recorded actions, and were asked to identify the agent (self vs. other). Manipulations were applied to the visual information (actions differing in complexity, and degradation from 15 to 2 point-lights within the same clapping action) and to the auditory information (selfgenerated vs. externally-generated vs. none). Results indicate that self-recognition was better than chance in all conditions and was highest when observing relatively unconstrained patterns of movement. Auditory information did not increase accuracy even with the most ambiguous visual displays, suggesting that judgments of agent identity depend much more on motor cues than on auditory (action-generated) or audiovisual (synchronization) information. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Most of the actions humans perform rely on multisensory information from the environment. For example, music performance and perception are activities that occur in multisensory contexts. Music making requires precise timing, coordination and motor control for planning and executing one’s own movements, and for predicting the intentions and actions of others when playing in ensembles (Keller, 2008). In order to act appropriately, concurrent information from different sensory modalities must be efficiently processed simultaneously in space and time. Music listening is often accompanied by spontaneous body movements, even in young children. In everyday situations, the synchronization of body movements with music is a common activity, even for people without formal training, whether they are dancing, marching, or simply clapping in time during a concert. Does this contextual multisensory information help a person to know about his or her actions and the actions of others, or is the body movement alone sufficient? This study examines whether the dynamic relationship between an individual’s movements and externally generated auditory information, as well as auditory information generated by the action itself, provide cues for self-recognition during the observation of impoverished visual displays of actions performed in synchrony with music. It has been proposed that people understand their own and others’ actions by means of action simulation, that is by mapping observed movements onto their own action system (Jeannerod, 2006; see also the related principle of common coding, Hommel, Müsseler, Aschersleben, & Prinz, 2001). Evidence from single cell recordings in macaque monkeys (Gallese, Fadiga,
* Corresponding author. Address: Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstrasse 1A, 04103 Leipzig, Germany. Fax: +49 341 99 40 113. E-mail address: [email protected] (V. Sevdalis). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.017
618
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
Fogassi, & Rizzolatti, 1996) and imaging studies of humans (Fadiga, Fogassi, Pavesi, & Rizzolatti, 1995; Grèzes, Armony, Rowe, & Passingham, 2003) suggests that there is a close correspondence between neural activity while observing, imagining or executing the same action (for a review, see Rizzolatti & Craighero, 2004). Furthermore, embodied simulation and embodied cognition accounts (Gallese, 2007; Wilson, 2002) claim that social cognition and action understanding are grounded in the behavioral matching and neuronal overlap between action execution and action perception. Behavioral evidence for a close relationship between perception and action has been obtained in a number of studies of self-recognition (for a review, see Knoblich, 2008). Self-recognition has been investigated mainly by focusing on either visual or auditory modalities. In the visual modality, designs employing observation of hand gestures (Daprati, Wriessnegger, & Lacquaniti, 2007), drawing movement trajectories (Knoblich & Prinz, 2001) and various full body movements (Loula, Prasad, Harber, & Shiffrar, 2005) have been used for investigating self-recognition. In the auditory modality, self-recognition has been examined by having individuals listen to the sounds of their own or others’ clapping (Flach, Knoblich, & Prinz, 2004) and by having piano players discriminate between their own and others’ musical performances (Repp & Knoblich, 2004). Studies of overt coordination with self- vs. other-generated stimuli (Flach, Knoblich, & Prinz, 2003; Keller, Knoblich, & Repp, 2007) have provided further support for the claim that the perception of agent identity is grounded in motor processes, specifically, in action simulation (Jeannerod, 2003, 2006). For example, in the music domain, Keller et al. (2007) found that pianists were not only able to recognize their own performances, but were also able to synchronize better when playing duets with their own previous recordings than with another pianist’s recordings. Thus, visual and auditory cues to agent identity are provided by the idiosyncratic ways in which individuals move due to personal biomechanical constraints and past experience and training. Neurophysiological and neuroimaging evidence suggests that the perception–action links are implemented on a neuronal level (Blakemore & Decety, 2001; Decety & Grèzes, 1999; Jeannerod, 2001; Rizzolatti & Craighero, 2004). In the domains of music (Bangert & Altenmüller, 2003) and dance (Calvo-Merino, Glaser, Grezes, Passingham, & Haggard, 2005; Cross, Hamilton, & Grafton, 2006; Cross, Kraemer, Hamilton, Kelley, & Grafton, 2009), it has been shown that perception–action links become more robust as a function of training and expertise, leading to stronger activation of brain areas that are related to action planning and motor performance in experts relative to novices (Haslinger et al., 2005; Haueisen & Knösche, 2001). Additional evidence for shared auditory and motor processing networks in musical activities comes from studies by Lahav, Saltzman, and Schlaug (2007), Lindenberger, Li, Gruber, and Muller (2009), Mutschler et al. (2007) and Zatorre, Chen, and Penhune (2007). Self-recognition studies employ various methods, paradigms and designs, including online (concurrent) vs. offline (delayed) tasks, full body movements vs. fine movements of body effectors and goal-directed vs. nongoal-directed actions. Furthermore, self-recognition has been studied by using point-light displays (Loula et al., 2005), the rubber hand illusion paradigm (Botvinick & Cohen, 1998), neuroimaging (for reviews see Legrand & Ruby, 2009; Lenggenhager, Smith, & Blanke, 2006) and more recently, virtual reality paradigms (Lenggenhager, Tadi, Metzinger, & Blanke, 2007). The main methodological technique in self-recognition experiments is to observe and compare self- vs. other-generated action signals. The crucial question is how signals that stem from oneself or another person are monitored and used in order to disambiguate the identity of the bodies and the origin of the actions. The interplay between efferent and afferent signals is considered a key factor in the sense of agency – the experience that oneself is the cause of an ongoing action (Gallagher, 2000, 2005). Recent evidence suggests that the sense of agency may depend especially on efferent (motor) signals from one’s own actions (Engbert, Wohlschläger, & Haggard, 2008; Tsakiris & Haggard, 2005; Tsakiris, Haggard, Franck, Mainy, & Sirigu, 2005). Although efferent information can modulate awareness of an action’s timing, the sensory processing of self-generated events and action attribution (Tsakiris & Haggard, 2005), the role of afferent signals in the conscious phenomenal experience of agency remains unclear (Tsakiris, 2008). The meaning of afferent signals for perception is ambiguous because afferent signals may be either self- or externally-generated (i.e., reafferent or ex-afferent; see also Knoblich & Repp, 2009; Repp & Knoblich, 2007). However, if the pattern of afferent information is such that it seems familiar or evokes resonance in the observer’s motor system, it may be sufficient for self-recognition. Most of the studies that aimed at investigating the role of motor cues in on-line judgments of agency (e.g., Daprati et al., 1997; Sirigu, Daprati, Pradat-Diehl, Franck, & Jeannerod, 1999; Van den Bos & Jeannerod, 2002) have used simple actions of specific body effectors (e.g., finger flexion, hand tapping). By contrast, studies of off-line self-recognition often have used more complex, whole-body actions. Even when human action is depicted by just a few point-lights (Johansson, 1973; for a review, see Blake & Shiffrar, 2007), observers can identify themselves, their friends or strangers from these point-light movement trajectories (Beardsworth & Buckner, 1981; Cutting & Kozlowski, 1977; Jokisch, Daum, & Troje, 2006; Loula et al., 2005; Prasad & Shiffrar, 2009). When walking movements were used, participants were able to recognize themselves and their friends equally well (Cutting & Kozlowski, 1977) and there was only a small advantage for self vs. friend recognition in a study by Beardsworth and Buckner (1981). In both studies, however, recognition rates were barely above chance. In the Loula et al. (2005) study, where a number of different actions were displayed, self-recognition was better than recognition of a friend’s movements, although agent recognition was still not above chance for some actions such as walking. Recognition accuracy was action dependent, with the rich kinematic information contained in actions such as dancing affording a higher degree of recognition. In the Jokisch et al. (2006) study, distinction of self and other in walking patterns was achieved, although recognition for other’s actions deteriorated when the actions were seen in profile rather than in a frontal viewpoint. By applying viewpoint manipulations, Prasad and Shiffrar (2009) obtained results that led them to conclude that
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
619
visual experience with one’s own movements is of less importance in self-identification, which supports the hypothesis that motor resonance in action perception enables self-recognition. The current study investigates self-recognition of actions performed by non-musicians in synchrony with music. Coordination with music may carry information about personal action styles in the way in which movements are timed relative to the music’s rhythmic structure. Listening and/or moving to music may be like interacting with a virtual partner (Leman, 2007), and this interaction may be at the core of musical experience (Overy & Molnar-Szakacs, 2009). In research conducted outside the music domain, contextual cues (e.g., scenes surrounding the action, presence of another agent) have been previously shown to have an impact on the way actions are both perceived and performed (Iacoboni et al., 2005; Sebanz, Knoblich, & Prinz, 2005). By implementing a research paradigm that combines visual (point-light) displays of full body movements with auditory information (self- or externally-generated), we aim at elucidating the influence of contextual multisensory cues on judgments of agent identity, which have been previously studied in either the visual or the auditory modality alone. Our goal is therefore to assess self-recognition in more ecologically valid, multisensory contexts, which are typical for music-related behaviors. Furthermore, the actions considered are representative of realistic behaviors that vary in complexity or, more specifically, in the degrees of freedom that constrain movement kinematics: dancing, walking, and clapping in time with music. We hypothesized that, in addition to intrinsic information that stems from an agent’s movements in space and time, extrinsic information about the relation of movements to the music may potentially affect self-recognition performance. In other words, by examining self-recognition in point-light action displays presented together with auditory information, we test the hypothesis that contextual information about synchrony between movements and music improves recognition relative to information about movement kinematics alone. Specifically, music provides a temporal reference frame against which movement timing can be gauged. If such information were to contribute to self-recognition, it would be most likely to do so with relatively constrained actions (walking and clapping) whose agents were often not reliably identified in previous studies. This would be consistent with the inverse effectiveness principle (Alais & Burr, 2004; Meredith & Stein, 1983), which states that an observer’s perception can compensate for the reduced quality of a stimulus in one modality by making use of information in another modality. Based on results of the study by Loula et al. (2005) and basic psychophysical premises (e.g., Weber–Fechner law), we assumed that recognition would be at a high level for dancing even in the visual-only condition, leaving little room for boosted performance due to additional sources (e.g., synchronization cues). 2. Experiment 1 The main aim of Experiment 1 was to examine whether contextual multisensory information has an impact on self-recognition. With regard to visual information, based on the results of previous research (Loula et al., 2005), we expected that self-other discrimination would improve with decreasing constraints on movement, and, therefore, would be most accurate for dancing and least accurate for clapping. With regard to contextual information, we wanted to examine whether concurrent auditory information about the dynamic relationship between an individual’s movements and music provides informative cues for self-recognition. 2.1. Method 2.1.1. Participants Fourteen adults (8 females; aged 21–29; mean age: 24.2 years) participated in the study in return for financial compensation. All of them reported to have normal hearing and normal or corrected to normal vision. None of them had previous experience with point-light displays. Participants did not know each other and were not informed about the experimental hypotheses. 2.1.2. Design The experiment was carried out in two sessions, in a within subjects design, separated by 1–2 months. The sessions were named ‘action session’ and ‘perception session’. The independent variables of interest in the perceptual experiment were action (dancing, walking, clapping) and modality (presence vs. absence of auditory information). 2.1.3. Action session 2.1.3.1. Materials. Excerpts from three musical pieces of different styles (‘drum and bass’ dance, jazz and folk) were used as stimuli. The pieces were ‘Apache’, in a version performed by the Incredible Bongo Band, ‘In the Mood’ by the Glenn Miller Orchestra, and ‘Madan’ by Salif Keita. The excerpts were in .wav format, and their duration was 60–70 s. The excerpts had a clear beat and were selected on the basis that they would represent a variety of genres and that their tempi would allow the execution of all the three actions (clapping, walking and dancing) at a comfortable rate. 2.1.3.2. Equipment and procedure. Participants executed each of the three different actions in time with each of the three musical excerpts. Performances were recorded by a Vicon motion capture system (Vicon – Oxford, UK). Ten cameras were placed at relatively equal distances from the center of the room. Data acquisition with a sampling frequency of 200 fps
620
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
Fig. 1. Point-light depiction of an agent with 13 markers attached at the head and the main joints of the body.
was controlled by Vicon Nexus software. At the start of each musical excerpt, a digital signal was sent to a trigger panel, thus enabling the precise onset time of the audio signal to be recorded along with the motion capture data. Performances were also videotaped by a digital video camera (SONY HDR-HC9). Participants came to the lab individually and were told that they would be required to clap, walk and dance in synchrony with the beat of three musical excerpts and that their movements would be recorded. They were informed that this experiment was not a test to judge their dancing abilities and that they should simply listen to the music and execute the actions in time to it. Music was used in order to exercise control over the timing of action execution. If participants were invited to clap or walk without an external referent (i.e., the beat of music), then their actions would presumably be much more diverse and variable in tempo. The treatment of the participants complied with all the ethical guidelines of the Max Planck Institute for Human Cognitive and Brain Sciences, where the research was conducted. All participants signed a consent form for their participation before the experiment begun. Thirteen reflective markers were attached to the participants’ bodies, at the head and at the main joints (shoulders, elbows, wrists, hips, knees and ankles) (see Fig. 1). For the clapping action, one extra marker was attached on the proximal phalanx of each index finger to enhance visibility and clarity of the clapping action. Participants first listened to one musical excerpt to become familiar with it. Then they executed the three different actions, one at a time, for the entire duration of the excerpt. The desired time intervals of claps and walking steps were indicated by an experimenter and were the same for all the participants. The actions took place within the same starting positions and walking paths, which were clearly indicated by white tape on the floor (clapping: clapping while standing at the center of the room; walking: walking 5.50 m from the one end of the room to the other and then back; dancing: starting position at the center of the room and available dancing space of 280 280 cm). After all the actions had been executed for the first musical excerpt, the same procedure was repeated for the second and the third musical excerpts. The orders of the excerpts and of the actions were randomized. The duration of the action session did not exceed 1 h.1 2.1.4. Perception session 2.1.4.1. Materials. Point-light movies were prepared using Final Cut Pro and QuickTime Software. The onset times of motion capture and audio data were synchronized by aligning the motion capture files of the agents in frontal view with the motion capture files of the trigger panel onset signals and with the audio files. From each combination of 3 actions 3 musical excerpts, two 5-s excerpts were randomly selected and served as stimuli. The point-light movies displayed the performances of the three different actions (dancing, walking, and clapping) with or without the accompanying music (Audiovisual vs. Visual condition). Participants were matched in pairs according to gender and physical body proportions (i.e., for every ‘self’ there was just one ‘other’). The matching process was carried out in the same way as in previously published studies (Loula et al., 2005). Matching participants according to gender and physical body proportions ensured that neither gender (Kozlowski & Cutting, 1977) nor weight (Runeson & Frykholm, 1983) could be used as the basis for participants’ discriminations. Thus, seven matched pairs (4 female pairs and 3 male pairs) were created. In total, for every pair, 72 point-light movies were created that contained all the possible combinations of agents, actions, modalities, musical pieces, and different episodes of each action (2 agents 3 actions 2 auditory conditions 3 pieces 2 selections). From these movies, 36 movies depicted the ‘self’ and 36 movies depicted the ‘other’. Additionally, 16 extra movies were created as practice trials (8 for ‘self’ and 8 for ‘other’). The 5-s excerpts selected for practice trials were different from those in test trials.
1 At the end of the session, participants were required to rate the familiarity of the musical pieces on a scale from 1 to 3 (anchors: unfamiliar–familiar). The dance piece received an average familiarity rating of 1.43 (SD = 0.76), the jazz piece 2.57 (SD = 0.76) and the folk piece 1.36 (SD = 0.63). In all subsequent analyses, no significant differences in performance were found across pieces and familiarity did not correlate with performance.
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
621
Fig. 2. Average d0 scores in the two modality conditions for the three actions. Error bars indicate 95% confidence intervals.
2.1.4.2. Equipment and procedure. The same participants were invited back and watched the 5-s point-light movies on a computer monitor. The point-light figures were projected at 11.42 degrees of visual angle, their height was 10 cm and the viewing distance was about 50 cm for all the participants. The session started with the participants watching a block of eight practice trials (4 for ‘self’ and 4 for ‘other’, randomly selected from the 16 practice trials). Then each participant watched 144 point-light movies presented in random order (36 ‘self’ and 36 ‘other’ movies repeated across two blocks). Before each trial onset, a white fixation cross appeared at the center of the monitor, lasting for 1-s. The auditory information was delivered over two loudspeakers placed to the left and right in front of the participants, at a distance of 2 m. The task was to indicate by key press on a computer keyboard whether the depicted agent was oneself or another person. On the keyboard, the left arrow was labeled ‘S’ for ‘selbst’, indicating ‘self’, and the right arrow was labeled ‘A’ for ‘andere’, indicating ‘other’. No feedback about correctness was provided after responses. The duration of the perception session did not exceed 1 h. 2.2. Results and discussion Self-other discrimination was assessed by computing d0 . This measure takes response bias into account by subtracting ztransformed false alarm rates (‘self’ judgments for ‘other’ displays) from hit rates (correct ‘self’ responses) (see Macmillan & Creelman, 1991). The results are shown in Fig. 2. Recognition accuracy was significantly better than chance (d0 = 0) for all actions: dance, t(13) = 22.60, p < .001; walk, t(13) = 6.25, p < .001; clap, t(13) = 3.54, p < .01. In raw scores, when collapsed across all conditions and stimuli (modality conditions, musical pieces, and selections), correct recognition performance (i.e., ‘self’ as ‘self’ and ‘other’ as ‘other’) for the dancing action was 94.49% (SE = 3.09), for the walking action 79.46% (SE = 7.26), and for the clapping action 76.04% (SE = 9.28). A 3 2 repeated measures analysis of variance (ANOVA) was conducted on d0 scores to test our hypotheses about the effects of Action (dance vs. walk vs. clap) and Modality (visual vs. audiovisual) on self-recognition. This analysis revealed a statistically significant main effect of Action, F(2, 26) = 3.36, p = .05, g2p ¼ :21. Paired t-tests revealed that the difference in performance between the dance and walk conditions was significant, t(13) = 3.91, p < .01, while the difference between the walk and clap conditions was not significant, t(13) < 1, n.s. The main effect of Modality was not significant in the ANOVA, indicating no reliable differences in self-recognition between audiovisual and visual conditions, F(1, 13) < 1, n.s. The interaction between Action and Modality was likewise not significant, F(2, 26) < 1, n.s. To summarize, recognition accuracy was highest for the most unconstrained movement patterns (i.e., for dancing), although agents were still recognized reliably in the case of the simplest action (clapping). The audiovisual presentation of point-light action patterns did not influence agent recognition for any action. Thus, the availability of additional cues about synchrony between movements and music did not improve recognition. 3. Experiment 2 The finding that auditory information did not boost performance in Experiment 1 suggests either that visual information about the action kinematics was sufficient for self-recognition or that there was no information to be gained from the relationship between the movements and the music. The information in the visual displays was quite rich because 13 pointlights (dancing and walking) or 15 point-lights (clapping) were used. Therefore, in the current experiment we tested whether agent recognition is still possible when parametric degradation is applied to the visual information. We hypothesized that auditory information might improve self-recognition when the visual information is severely impoverished. Furthermore, another reason why auditory information did not boost performance in Experiment 1 may have been the fact that no action-specific auditory information was provided. Therefore, we added a condition in Experiment 2 in which another kind of auditory information was added, one that was specific to the actions (clapping sounds). Clapping sounds have been shown in past studies to contain cues to agent identity (Flach et al., 2004). Thus we now also tested the hypothesis
622
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
that self-generated sounds could serve as ancillary cues to agent identity, particularly when visual information is degraded. Clapping was the only action investigated in Experiment 2. 3.1. Method 3.1.1. Participants Twelve out of the fourteen adults that participated in Experiment 1 could be contacted and were invited back to participate in Experiment 2, which took place 2–3 months later. 3.1.2. Design The independent variables were the visual information available in the clapping displays (original full-body condition and three degradation levels) and the additional auditory information provided (none, music only, and music + clapping sounds). 3.1.3. Materials The clapping performances from the action session of Experiment 1 were used to prepare stimuli for Experiment 2. We introduced three degradation levels to the visual information available. In the degraded stimuli, only the body effectors that were involved in the execution of the clapping action were retained (i.e., the arms) and all the other markers were discarded. This degradation process resulted in four visual conditions (see Fig. 3): full body (15 markers), shoulders–elbows–wrists–fingers (8 markers), elbows–wrists–fingers (6 markers), and fingers only (2 markers). Auditory information was also manipulated to produce two auditory conditions in addition to a silent condition. One condition contained only the music (dance, folk or jazz) that initially accompanied the action (Music) and the other contained the sounds of each participant’s clapping along with the music (Music + Clap). From each participant’s clapping action, three 5-s excerpts were randomly selected to serve as point-light movies. In order to reduce the familiarity with the stimuli, these selected excerpts were different than those used in Experiment 1. The same matched participant pairs as in the previous experiment were used. In total, for every pair, 216 new point-light movies were created that contained all the possible combinations of agents, visual degradation conditions, auditory conditions, musical pieces, and different episodes of each action (2 agents 4 degradation levels 3 auditory conditions 3 pieces 3 selections). From these movies, 108 movies depicted the ‘self’ and 108 movies depicted the ‘other’.
Fig. 3. Point-light depiction of an agent with 15, 8, 6 and 2 markers attached at the head, the main joints, and the fingers.
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
623
Fig. 4. Average d0 scores in the three modality conditions for the four degradation levels. Error bars indicate 95% confidence intervals.
3.1.4. Equipment and procedure The experiment started with the participants watching a block of 8 practice trials, selected randomly from 16 practice trials. In total, each participant watched 216 point-light movies on a computer monitor, presented in random order (108 ‘self’ and 108 ‘other’ movies). All other aspects of the procedure were as in Experiment 1. 3.2. Results and discussion Self-other discrimination was again assessed by computing d0 . The results are shown in Fig. 4. Two participants were excluded from the analyses due to failure to comply with the experimental instructions.2 For the remaining participants, recognition accuracy was significantly better than chance for all degradation levels: ts(9) > 4.14, p < .01.3 In raw scores, when collapsed across all other conditions, correct recognition performance with 15 markers was 91.30% (SE = 6.26), with 8 markers 87.04% (SE = 7.55), with 6 markers 85.00 (SE = 8.14), and with 2 markers 78.70% (SE = 10.29). A 4 3 repeated measures analysis of variance (ANOVA) was conducted on the d0 scores to test our hypotheses about the effects of visual Degradation (15 vs. 8 vs. 6 vs. 2 markers) and auditory information (none, music, music + clap) on self-recognition. This analysis revealed only a statistically significant main effect of Degradation, F(3, 27) = 6.25, p < .01, g2p ¼ :41. A significant linear trend across degradation levels confirmed the reliability of the decrease in recognition from 15 to 2 markers, F(1, 9) = 10.32, p = .01. The main effect of auditory information was not significant in the ANOVA, F(2, 18) < 1, n.s., and the interaction was likewise not significant, F(6, 54) < 1, n.s. To summarize, recognition accuracy for clapping agents decreased with decreasing availability of kinematic information, although agents could still be recognized reliably in the case of highly degraded displays (2 markers). Furthermore, agent recognition was reliable irrespective of whether self-generated auditory information (clapping sounds) or externally-generated sound (music) accompanied the actions. Therefore, self-other discrimination was achievable based only on kinematic cues, without any additional support from auditory cues. 4. General discussion The aim of this study was to investigate the relative contributions of multisensory spatial–temporal cues that can be used for explicit judgments of agent identity. Our experiments show that self-recognition was based exclusively on visual information about personal movement kinematics, with contextual cues about synchrony between movements and sounds conferring no discernable benefit, even in impoverished visual conditions. The current findings extend research on the links between action execution and action perception to actions performed in synchrony with music. The sensitivity of the action-observation network may depend on the nature of the action: Relatively unconstrained movements and spatially unoccluded actions may be particularly potent due to richness in visual information about personal styles of action execution.
2 These participants had very large, negative values of d0 , while all other participants had positive values of d0 . This deviant performance indicates that they reversed their responses, responding ‘self’ when they should have responded ‘other’, and vice versa. Although we could have simply reversed these participants’ responses, because the absolute values of their discrimination sensitivity indices were high and hence their ability to distinguish agents per se was good, we considered it more prudent to exclude them. If the analyses are carried out with the reversed scores, all the effects reported above are maintained. The apparent difference in d0 scores for the clapping action between the Experiments 1 and 2 is due to the exclusion of these two participants in Experiment 2. 3 In order to test whether the better-than-chance performance observed with 2-marker displays was partially attributable to priming by earlier presented displays of the given action with more markers, we extracted data for all 2-marker trials that occurred before any trials with a greater number of markers, and then computed recognition accuracy for each participant based only on these data (note that this could not be done in the case of one participant, for whom all 2-marker displays occurred after the other conditions). This analysis revealed that participants’ self-other judgments were on average accurate for 73.3% of 2marker displays that were encountered before seeing fuller displays of the clapping action. This accuracy rate was significantly better than chance (50%) (t(8) = 6.16, p < .01).
624
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
Under such conditions, self-recognition may be driven more by visual than by auditory cues. The high informativeness of visual cues is evidenced by the fact that self-recognition was possible even when simple actions were depicted as point-light displays and even in highly degraded displays (i.e., displays with high spatial occlusion). Furthermore, it should be noted that the results obtained in our experiments concern actions that were not directed towards objects or goals. Finally, participants in our study performed the discrimination task with minimal training, without any feedback, and without having any previous experience with point-light displays. The actions used in our experiments are gross motor movements (dancing, walking and clapping), which may afford easier agent recognition in comparison to previous experiments that focused on fine motor movement recognition (Daprati et al., 1997, 2007; Sirigu et al., 1999; Van den Bos & Jeannerod, 2002). Furthermore, whereas dancing may enable larger individual differences, in both of our experiments, agents were recognized even when the ‘self’ and the ‘other’ were performing movements in a very similar way (clapping). Even the highly ambiguous displays (two point-lights clapping) may contain a sufficient amount of the information needed to recognize an agent. This information is provided by the agents’ movements in space and time and not by differing limb lengths and body proportions, as these anatomical features remained constant while self-other discrimination varied across different actions (Experiment1) and different versions of the same action (Experiment 2). The fact that agents were distinguishable between different actions (Experiment 1) and for different versions of the same action (Experiment 2) may also imply that self-related kinematic properties possess highly specified gestalts. Perceptual information can be extracted by particular configural properties specified in action patterns (see also Pinto & Shiffrar, 1999). The occluded elements can perhaps be ‘filled in’ in the mind of the observer, and, therefore, make the agent identification possible. This suggestion is consistent with previous studies (Loula et al., 2005), supporting the role of action simulation for self-recognition. Recognition may be based on visual perception of idiosyncratic information, patterned in specified action gestalts. Minimal visual information has been shown to be sufficient for action perception in a wide variety of music-related and other contexts. In the music domain, it has been shown that inexpressive body movements (Davidson, 1993) and reductions in visual information (Davidson, 1994) may provide sufficient cues for detecting a performer’s intentions when presented as point-light displays. Furthermore, performer’s intentions can be detected from blurred video images of specific body effectors (Dahl & Friberg, 2007). In other domains, Kozlowski and Cutting (1977) have reported that minimal visual information is sufficient for identifying the gender of a point-light actor and Pollick, Paterson, Bruderlin, and Sanford (2001) reported that arm movements are informative about the agent’s underlying affective state. Movement kinematics may be considered in Gibson’s terms (Gibson, 1979) as invariant structures that can specify agent identity. Even under audiovisual transformations of information, the core elements of the sense of being an action’s agent remain relatively intact and stable over time. The kinematic specification of dynamics (KSD) principle (Runeson & Frykholm, 1983) may account for the fact that the movement kinematics in point-light displays, apart from specifying action properties, expectancies and intentions, may also specify person-related individual characteristics. Therefore, if person-related dynamics may be identified from movement kinematics alone, then the availability of further dynamic information that accompanies the action (synchronization with music) or specifies the action (clap sounds) does not have an additive effect on identification. In other words, extra dynamic cues added to person-related dynamic properties may be redundant and thus of no use for the perceptual judgments. In Brunswik’s terms, the information that specifies the actions of the self (i.e., the cues available) is a source of noise or uncertainty. The task of the organism is to base its perceptual judgments on cue-based probabilistic inferences of the information available (Brunswik, 1956; Hammond & Stewart, 2001). In our experiments, during the inferential process of the ‘self’ identifying ‘self’ or ‘other’ based on visual and auditory signals, some cues are given higher perceptual weights and some may be ignored. When multiple cues were available for recognizing self vs. other, participants relied on the kinematic information. In other words, kinematic information was sufficient for self-recognition. This implies that judgments of self-recognition depend on the availability and reliability of the cues provided. Different combinations of cues may lead to different strategies that determine the weighting of sensory and motor signals. The current results suggest that, when combined with visual cues, contextual auditory cues were superseded. This finding may imply a hierarchy of cues for self-recognition. In our experiments, visually communicated kinematic information seems to be at the top of this hierarchy. Why do participants ignore auditory information? Firstly, according to the modality appropriateness hypothesis (Welch & Warren, 1980), discrepancies between modalities are resolved in favor of the most precise or the most appropriate modality for the given task or situation. In our experimental task, visual information may have dominated over information from other modalities. Moreover, the current results seem to challenge the inverse effectiveness principle (Alais & Burr, 2004; Meredith & Stein, 1983). In our experiments, the reduction of spatial information available in the visual modality did not increase reliance upon additional auditory information (whether self- or externally-generated). The ability to detect self vs. other information from unimodal (soundless) point-light displays may suggest that the comparisons for agent identity across modality conditions (visual vs. audiovisual) are made using a modality-neutral metric. Judgements of agent identity may be reliable even if they are executed based on unimodal information. Furthermore, temporal information may be encoded and embodied motorically when biological motion is perceived. Auditory cues may be ignored when visually observing the self in action potentially because action cues may modulate auditory and visual information perception. Another explanation may be the robustness of the action signals per se. Kinematic patterns may provide salient information for efficient self-other discrimination, thus, rendering all other information redundant. The current findings suggest that self-recognition may be an inferential process determined by an optimal combination of the perceptual weights allocated to the cues available.
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
625
Despite the absence of effects of auditory information in our study, the role of audiovisual cues in the perception of agent identity in musical contexts deserves further attention in future research. For instance, it may be fruitful to assess how selfrecognition is affected by systematic manipulations of the degree of temporal (in)congruence between movements and auditory information in terms of phase (lead/lag) and periodicity (tempo) relations. Such exploration of the potential cues that mediate perception and action may provide further insights about how humans understand the actions of the self and others. Communicated information is multimodally encoded and embedded in interactive environments. By investigating interaction settings from multiple perspectives, one may gain an understanding of the constituent mechanisms involved in interpersonal relations and identify how these mechanisms enable shared intentions, joint actions and cooperation. Acknowledgments The research was supported by the Max Planck Society. The authors would like to thank Bruno Repp for comments on an earlier version of this paper, Jan Bergmann for programming and technical assistance, Kerstin Träger for technical support, and Regine Steinke and Juliane Zeiss for their help with data collection. The results of Experiment 1 are reported in brief form in the proceedings of The Neurosciences and Music – III (Sevdalis & Keller, 2009). References Alais, D., & Burr, D. (2004). The ventriloquism effect results from near-optimal bimodal integration. Current Biology, 14, 257–262. Bangert, M., & Altenmüller, E. O. (2003). Mapping perception to action in piano practice: A longitudinal DC-EEG study. BMC Neuroscience, 4, 26. Beardsworth, T., & Buckner, T. (1981). The ability to recognize oneself from a video recording of one’s movements without seeing one’s body. Bulletin of the Psychonomic Society, 18, 19–22. Blake, R., & Shiffrar, M. (2007). Perception of human motion. Annual Review of Psychology, 58, 47–73. Blakemore, S., & Decety, J. (2001). From the perception of action to the understanding of intention. Nature Reviews Neuroscience, 2, 561–567. Botvinick, M., & Cohen, J. (1998). Rubber hands ‘feel’ touch that eyes see. Nature, 391, 756. Brunswik, E. (1956). Perception and the representative design of psychological experiments. Berkeley: University of California Press. Calvo-Merino, B., Glaser, D. E., Grezes, J., Passingham, R. E., & Haggard, P. (2005). Action observation and acquired motor skills: An fMRI study with expert dancers. Cerebral Cortex, 15, 1243–1249. Cross, E. S., Hamilton, A. F. D. C., & Grafton, S. T. (2006). Building a motor simulation de novo: Observation of dance by dancers. Neuroimage, 31, 1257–1267. Cross, E. S., Kraemer, D. J. M., Hamilton, A. F. D. C., Kelley, W. M., & Grafton, S. T. (2009). Sensitivity of the action observation network to physical and observational learning. Cerebral Cortex, 19, 315–326. Cutting, J. E., & Kozlowski, L. T. (1977). Recognizing friends by their walk – Gait perception without familiarity cues. Bulletin of the Psychonomic Society, 9, 353–356. Dahl, S., & Friberg, A. (2007). Visual perception of expressiveness in musicians’ body movements. Music Perception, 24, 433–454. Daprati, E., Franck, N., Georgieff, N., Proust, J., Pacherie, E., Dalery, J., et al (1997). Looking for the agent: An investigation into consciousness of action and self-consciousness in schizophrenic patients. Cognition, 65, 71–86. Daprati, E., Wriessnegger, S., & Lacquaniti, F. (2007). Kinematic cues and recognition of self-generated actions. Experimental Brain Research, 177, 31–44. Davidson, J. W. (1993). Visual perception of performance manner in the movements of solo musicians. Psychology of Music, 21, 103–113. Davidson, J. W. (1994). What type of information is conveyed in the body movements of solo musician performers? Journal of Human Movement Studies, 6, 279–301. Decety, J., & Grèzes, J. (1999). Neural mechanisms subserving the perception of human actions. Trends in Cognitive Sciences, 3, 172–178. Engbert, K., Wohlschläger, A., & Haggard, P. (2008). Who is causing what? The sense of agency is relational and efferent-triggered. Cognition, 107, 693–704. Fadiga, L., Fogassi, L., Pavesi, G., & Rizzolatti, G. (1995). Motor Facilitation during action observation – A magnetic stimulation study. Journal of Neurophysiology, 73, 2608–2611. Flach, R., Knoblich, G., & Prinz, W. (2003). Off-line authorship effects in action perception. Brain and Cognition, 53, 503–513. Flach, R., Knoblich, G., & Prinz, W. (2004). Recognizing one’s own clapping: The role of temporal cues. Psychological Research, 69, 147–156. Gallagher, S. (2000). Philosophical conceptions of the self: Implications for cognitive science. Trends in Cognitive Sciences, 4, 14–21. Gallagher, S. (2005). How the body shapes the mind. Oxford: Oxford University Press. Gallese, V. (2007). Before and below ‘theory of mind’: Embodied simulation and the neural correlates of social cognition. Philosophical Transactions of the Royal Society B: Biological Sciences, 362, 659–669. Gallese, V., Fadiga, L., Fogassi, L., & Rizzolatti, G. (1996). Action recognition in the premotor cortex. Brain, 119, 593–609. Gibson, J. J. (1979). The ecological approach to visual perception. Boston, MA: Houghton Mifflin. Grèzes, J., Armony, J. L., Rowe, J., & Passingham, R. E. (2003). Activations related to ‘‘mirror” and ‘‘canonical” neurones in the human brain: An fMRI study. Neuroimage, 18, 928–937. Hammond, K. R., & Stewart, T. R. (Eds.). (2001). The essential Brunswik: Beginnings, explications, applications. New York: Oxford University Press. Haslinger, B., Erhard, P., Altenmüller, E., Schroeder, U., Boecker, H., & Ceballos-Baumann, A. O. (2005). Transmodal sensorimotor networks during action observation in professional pianists. Journal of Cognitive Neuroscience, 17, 282–293. Haueisen, J., & Knösche, T. R. (2001). Involuntary motor activity in pianists evoked by music perception. Journal of Cognitive Neuroscience, 13, 786–792. Hommel, B., Müsseler, J., Aschersleben, G., & Prinz, W. (2001). The theory of event coding (TEC): A framework for perception and action planning. Behavioral and Brain Sciences, 24, 849–878. Iacoboni, M., Molnar-Szakacs, I., Gallese, V., Buccino, G., Mazziotta, J. C., & Rizzolatti, G. (2005). Grasping the intentions of others with one’s own mirror neuron system. PLoS Biology, 3, 529–535. Jeannerod, M. (2001). Neural simulation of action: A unifying mechanism for motor cognition. Neuroimage, 14, S103–S109. Jeannerod, M. (2003). The mechanism of self-recognition in humans. Behavioural Brain Research, 142, 1–15. Jeannerod, M. (2006). Motor cognition: What actions tell the self. New York: Oxford University Press. Johansson, G. (1973). Visual perception of biological motion and a model for its analysis. Perception & Psychophysics, 14, 201–211. Jokisch, D., Daum, I., & Troje, N. F. (2006). Self recognition versus recognition of others by biological motion: Viewpoint-dependent effects. Perception, 35, 911–920. Keller, P. E. (2008). Joint action in music performance. In F. Morganti, A. Carassa, & G. Riva (Eds.), Enacting intersubjectivity: A cognitive and social perspective on the study of interaction (pp. 205–221). Amsterdam: IOS. Keller, P. E., Knoblich, G., & Repp, B. H. (2007). Pianists duet better when they play with themselves: On the possible role of action simulation in synchronization. Consciousness and Cognition, 16, 102–111. Knoblich, G. (2008). Bodily and motor contributions to action perception. In R. L. Klatzky, B. MacWhinney, & M. Behrmann (Eds.), Embodiment, ego-space and action (pp. 45–78). New York: Psychology Press.
626
V. Sevdalis, P.E. Keller / Consciousness and Cognition 19 (2010) 617–626
Knoblich, G., & Prinz, W. (2001). Recognition of self-generated actions from kinematic displays of drawing. Journal of Experimental Psychology: Human Perception and Performance, 27, 456–465. Knoblich, G., & Repp, B. (2009). Inferring agency from sound. Cognition, 111, 248–262. Kozlowski, L. T., & Cutting, J. E. (1977). Recognizing the sex of a walker from a dynamic point-light display. Perception & Psychophysics, 21, 575–580. Lahav, A., Saltzman, E., & Schlaug, G. (2007). Action representation of sound: Audiomotor recognition network while listening to newly acquired action. Journal of Neuroscience, 27, 308–314. Legrand, D., & Ruby, P. (2009). What is self-specific? Theoretical investigation and critical review of neuroimaging results. Psychological Review, 116, 252–282. Leman, M. (2007). Embodied music cognition and mediation technology. Cambridge, MA: MIT Press. Lenggenhager, B., Smith, S. T., & Blanke, O. (2006). Functional and neural mechanisms of embodiment: Importance of the vestibular system and the temporal parietal junction. Reviews in the Neurosciences, 17, 643–657. Lenggenhager, B., Tadi, T., Metzinger, T., & Blanke, O. (2007). Video ergo sum: Manipulating bodily self-consciousness. Science, 317, 1096–1099. Lindenberger, U., Li, S.-C., Gruber, W., & Muller, V. (2009). Brains swinging in concert: Cortical phase synchronization while playing guitar. BMC Neuroscience, 10, 22. Loula, F., Prasad, S., Harber, K., & Shiffrar, M. (2005). Recognizing people from their movement. Journal of Experimental Psychology: Human Perception and Performance, 31, 210–220. Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A user’s guide. Cambridge: Cambridge University Press. Meredith, M. A., & Stein, B. E. (1983). Interactions among converging sensory inputs in the superior colliculus. Science, 221, 389–391. Mutschler, I., Schulze-Bonhage, A., Glauche, V., Demandt, E., Speck, O., & Ball, T. (2007). A rapid sound–action association effect in human insular cortex. PLoS ONE, 2, e259. Overy, K., & Molnar-Szakacs, I. (2009). Being together in time: Musical experience and the mirror neuron system. Music Perception, 26, 489–504. Pinto, J., & Shiffrar, M. (1999). Subconfigurations of the human form in the perception of biological motion displays. Acta Psychologica, 102, 293–318. Pollick, F. E., Paterson, H. M., Bruderlin, A., & Sanford, A. J. (2001). Perceiving affect from arm movement. Cognition, 82, B51–B61. Prasad, S., & Shiffrar, M. (2009). Viewpoint and the recognition of people from their movements. Journal of Experimental Psychology: Human Perception and Performance, 35, 39–49. Repp, B. H., & Knoblich, G. (2004). Perceiving action identity – How pianists recognize their own performances. Psychological Science, 15, 604–609. Repp, B. H., & Knoblich, G. (2007). Toward a psychophysics of agency: Detecting gain and loss of control over auditory action effects. Journal of Experimental Psychology: Human Perception and Performance, 33, 469–482. Rizzolatti, G., & Craighero, L. (2004). The mirror-neuron system. Annual Review of Neuroscience, 27, 169–192. Runeson, S., & Frykholm, G. (1983). Kinematic specification of dynamics as an informational basis for-person-and-action perception: Expectation, gender recognition, and deceptive intention. Journal of Experimental Psychology: General, 112, 585–615. Sebanz, N., Knoblich, G., & Prinz, W. (2005). How two share a task: Corepresenting stimulus–response mappings. Journal of Experimental Psychology: Human Perception and Performance, 31, 1234–1246. Sevdalis, V., & Keller, P. E. (2009). Self-recognition in the perception of actions performed in synchrony with music. Annals of the New York Academy of Sciences, 1169, 499–502. Sirigu, A., Daprati, E., Pradat-Diehl, P., Franck, N., & Jeannerod, M. (1999). Perception of self-generated movement following left parietal lesion. Brain, 122, 1867–1874. Tsakiris, M. (2008). The self–other distinction: Insights from self-recognition experiments. In F. Morganti, A. Carassa, & G. Riva (Eds.), Enacting intersubjectivity: A cognitive and social perspective to the study of interactions (pp. 149–163). Amsterdam: IOP Press. Tsakiris, M., & Haggard, P. (2005). Experimenting with the acting self. Cognitive Neuropsychology, 22, 387–407. Tsakiris, M., Haggard, P., Franck, N., Mainy, N., & Sirigu, A. (2005). A specific role for efferent information in self-recognition. Cognition, 96, 215–231. Van den Bos, E., & Jeannerod, M. (2002). Sense of body and sense of action both contribute to self-recognition. Cognition, 85, 177–187. Welch, R. B., & Warren, D. H. (1980). Immediate perceptual response to intersensory discrepancy. Psychological Bulletin, 88, 638–667. Wilson, M. (2002). Six views on embodied cognition. Psychonomic Bulletin & Review, 9, 625–636. Zatorre, R. J., Chen, J. L., & Penhune, V. B. (2007). When the brain plays music: Auditory–motor interaction in music perception and production. Nature Reviews Neuroscience, 8, 547–558.
Consciousness and Cognition 19 (2010) 627–635
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Short Communication
The role of reversal frequency in learning noisy second order conditional sequences Thomas Pronk *, Ingmar Visser Department of Psychology, University of Amsterdam, Roetersstraat 15, 1018 WB, Amsterdam, The Netherlands
a r t i c l e
i n f o
Article history: Received 8 January 2009 Available online 18 February 2010 Keywords: Implicit learning Unconscious knowledge Sequence learning Subjective measure Objective measure Process dissociation procedure Sequential reaction time task Reward Subjective threshold
a b s t r a c t The hallmark of implicit learning is that complex knowledge can be acquired unconsciously. The second order conditionals (SOCs) of Reed and Johnson (1994) were developed to be complex, and they are popular materials for implicit learning research. Recently, it was demonstrated that in a sequence made noisy (by combining two SOCs), shared features of the SOCs may be learned explicitly (Fu, Fu, & Dienes, 2008). What are these shared features? We hypothesized that low reversal frequency may play a significant role. We have varied reversal frequency, and discovered that reversal frequency affected response times, inclusion exclusion behavior, and recognition ratings. Not only does it appear to be important to distinguish implicit and explicit knowledge, but also to distinguish what the knowledge is of. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Two decades ago it was proposed that complex temporal sequences could be learned unconsciously (Nissen & Bullemer, 1987; Reber, 1989). This phenomenon, known as implicit learning, has been studied intensely ever since (e.g. Cleeremans & McClelland, 1991; Frensch, Buchner, & Lin, 1994; Jiménez, Méndez, & Cleeremans., 1996; Perruchet & Amorim, 1992; Shanks & Johnstone, 1999; Shanks & Perruchet, 2002; and many others). Learning sequential dependencies is a basic form of learning that forms a prerequisite to knowledge acquisition in many domains such as causal learning (Glymour, 2003; Shanks, Holyoak, & Medin, 1996), social interactions (Colman, 1995), evaluative learning (de Houwer, Baeyens, & Hendrickx, 1997) and language learning (McShane, 1991; Pinker, 1994). In a typical sequence learning experiment, participants are presented with a sequence of stimuli that they have to respond to by pressing the appropriate key. Participants are not made aware that the sequence of stimuli contains regularities or repeating patterns. When participants are transferred to a different sequence, response times show a marked increase. A crucial characteristic for learning to be called implicit is the absence of explicit knowledge, which is subsequently measured by tasks that resemble recognition or recall tasks in memory research (cf. Roediger, 1990, see Perruchet & Amorim, 1992, for an early example of a recall like task in sequence learning). Nissen and Bullemer (1987), and much of the sequence learning research following them, used pseudo random sequences (i.e. random sequences without repeating stimuli twice in a row) as transfer sequences to establish reaction time differences. Reed and Johnson (1994) were concerned that transfer to pseudo random sequences would not provide an optimal assessment of sequence learning because the learning effect could also, at least partially, be explained by simpler associative
* Corresponding author. Fax: +31 20 525 6104. E-mail addresses: [email protected] (T. Pronk), [email protected] (I. Visser). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2009.12.008
628
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
learning processes (see also Cohen, Ivry, & Keele, 1990). Hence, they developed so-called second order conditional (SOC) sequences in which ‘‘every location is completely determined by the previous two locations, whereas knowing the previous location alone provides no information regarding the next location” (Reed & Johnson, 1994, p. 586). Using SOCs, both as learning material and as transfer sequence, assures that any reaction time (RT) differences between training and transfer are due to learning of SOC features rather than to learning of (adjacent) stimulus–stimulus associations. Since their 1994 paper, the SOCs presented by Reed and Johnson have become the stimulus material of choice in sequence learning research (for instance Destrebecqz & Cleeremans, 2001; Destrebecqz et al., 2005; Kelly, Burton, Riedel, & Lynch, 2003; Rauch et al., 1995, Shanks & Johnstone, 1999; Shanks & Perruchet, 2002; Stöcker, Hoffmann, & Sebald, 2003; Vaquero, Jiménez, & Lupiáñez, 2006; Werheid, Zysset, Muller, Reuter, & von Cramon, 2003). 2. Reversal frequency and noisy SOCs Nissen and Bullemer (1987), and Reed and Johnson (1994) used different blocks of stimuli (i.e. learning and transfer blocks) to establish RT differences, recently many researchers use so-called noisy sequences. Rather than presenting two SOCs in different blocks, in noisy SOCs, at each trial there is a certain probability for the stimulus to be drawn from either the training (own) SOC or from the transfer (other) SOC (Fu et al., 2008; Jiménez, Vaquero, & Lupiáñez, 2006; Schvaneveldt & Gomez, 1998; Shanks, Wilkinson, & Channon, 2003; Vandenberghe, Schmidt, Fery, & Cleeremans, 2006; Wilkinson & Shanks, 2004). The noise level is the probability with which the stimulus is drawn from other. The aim of using noisy SOCs rather than deterministic ones is to prevent the formation of explicit knowledge of own by making own less reliable (Wilkinson & Shanks, 2004). However, as suggested by Fu et al. (2008), exposure to noisy SOCs may promote learning of shared features. The SOCs developed by Reed and Johnson (1994) indeed have features in common that distinguish them from a (pseudo) random sequence. One of these features is reversal frequency. A reversal occurs when the third location of a sequence of three, henceforth referred to as a triplet, is the same as the first (e.g. 131, 424). This stimulus characteristic has also been named an alternation (Cleeremans & McClelland, 1991), alternating repetition (Jiménez et al., 1996), or an n-2 repetition (Koch, Philipp, & Gade, 2006). Various researchers have acknowledged that reversals have a special status in sequence learning (Curran, Smith, DiFranco, & Daggy, 2001; Koch et al., 2006). Because Reed and Johnson (1994) considered reversals as (undesirably) salient, their SOCs each contain a single reversal.1 As suggested by Howard et al. (2004), it may be learned that reversals are rare. In a random sequence, on average, one in three triplets is a reversal (Vaquero et al., 2006). Compared to a random sequence, the sequences used by Reed and Johnson (1994) have a low reversal frequency. Low reversal frequency might be learned and then be used to discriminate triplets that belong to own and other from triplets that do not belong to own and other. Learning of low reversal frequency may be promoted by the use of noisy SOCs. With increasing noise, own becomes less reliable, while low reversal frequency remains equally reliable. Therefore, learning of reversal frequency (as opposed to learning of own) might be more likely in noisy SOCs than in deterministic SOCs. Wilkinson and Shanks (2004), and Fu et al. (2008) trained participants on noisy SOCs, and measured implicit and explicit knowledge with the inclusion exclusion task (Destrebecqz & Cleeremans, 2001). The inclusion exclusion task is an application of the process dissociation procedure (which was introduced by Jacoby, 1991) to sequence learning. The inclusion exclusion task consists of two parts: in the inclusion task, a participant is requested to replicate the sequence he was trained on. In the exclusion task he is requested to avoid replicating this sequence. If the participant has implicit and explicit knowledge of the SOC, he should be able to control expressing this knowledge, thus being able replicate the SOC in inclusion, and avoid replicating the SOC in exclusion. If the participant has implicit but no explicit knowledge, he should be unable to control expressing his knowledge. Therefore, when a participant replicates the SOC regardless of inclusion or exclusion instructions, the underlying knowledge could be regarded as less explicit and more implicit (Destrebecqz & Cleeremans, 2001). Via the inclusion exclusion task, Fu et al. (2008) concluded that participants may have acquired explicit knowledge of the features shared by own and other. What these shared features might be was left unspecified. We propose that a feature shared by own and other, of which Fu et al. found explicit knowledge, is reversal frequency. When trained on own and other, participants are trained on a sequence with a low reversal frequency. If participants generate a sequence with a low reversal frequency, they would generate a high number of own and other. If participants have explicit knowledge of the low reversal frequency, then they would be able to control expressing the low reversal frequency. This would result in a higher reversal frequency in the exclusion task than in the inclusion task. A higher reversal frequency would result in lower numbers of own and other. When classifying all possible triplets (excluding repetitions), into own and other, twelve triplets remain, which Fu et al. (2008) named neither. Own and other contain one reversal each, and neither contains 10 reversals. Fu et al. found, with a noise level of p = .75 (i.e. 75% own and 25% other), that more neither triplets were generated in exclusion than in inclusion. In the current study we test whether the increase in neither is solely an increase in reversal frequency. If so, then the shared feature of own and other that participants acquired explicit knowledge of might have been their low reversal frequency. The goal of
1
Note that it is impossible to construct an SOC sequence consisting of four stimulus locations that does not contain a single reversal.
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
629
the current research is to establish such effects of reversal frequency in learning noisy SOCs, both in performance measures (RTs) and in subsequent measures of explicit knowledge. Part of the current experiment was a replication of Fu et al. (2008). We trained participants on noisy SOCs that contained a single reversal. Another group of participants was exposed to noisy SOCs that contained four reversals. If own, other and neither each contain four reversals, one cannot discriminate between (any set of) SOCs based on reversal frequency. Thus, knowing the reversal frequency does not provide any benefit for learning (any set of) SOCs. We hypothesize that in SOCs with four reversals, learning of reversal frequency should not occur. If reversal frequency is the only feature shared by own and other that participants learn explicitly, then in the one-reversal conditions the numbers of own and other should decrease from inclusion to exclusion task, and in the four-reversal conditions they should not. In the current research, recognition ratings were used as a second measure of explicit knowledge. Whereas the inclusion exclusion task measures explicit knowledge as the ability to control expressing this knowledge, recognition ratings measure explicit knowledge as the ability to express metaknowledge (Destrebecqz & Peigneux, 2005). If participants perform above chance on an objective measure, such as response time in the serial response time (SRT) task, but participants claim that they are guessing on a subjective measure such as recognition rating, then participants show knowledge, but no metaknowledge. This principle was proposed by Cheesman and Merikle (1984), and applied to sequence learning as the guessing criterion by Dienes and Berry (1997). There are indications that explicit knowledge as control and explicit knowledge as metaknowledge are convergent, as illustrated by Destrebecqz and Cleeremans’s (2001) finding that in conditions where implicit learning took place according to the inclusion exclusion task, recognition ratings were at chance levels. 3. Method 3.1. Participants Participants were 78 psychology students from the University of Amsterdam, compensated with 7 euro or with study credit for their participation. Following the procedure of Fu et al. (2008), participants were informed that 20 euro could be won with high performance. Seven participants were excluded from the analyses.2 The remaining participants had the following characteristics: the mean age was 20.7 years (SD = 2.37), 26 were male and 45 were female, six participants were lefthanded. 3.2. Apparatus The visual stimuli were presented on a 17 TFT display, sound was administered through headphones. The visual stimulus was a circle shown at one of four locations. Locations 1, 2, 3, and 4 corresponded to response keys D, F, J and K on the computer keyboard. These keys were pressed with the middle and index fingers of the left hand, and the index and middle fingers of the right hand. 3.3. Stimuli In the one-reversal conditions we have used the SOCs developed by Reed and Johnson (1994). In these SOCs, own and other contain a single reversal. For the four-reversal conditions we constructed a set of SOCs that each contain four reversals. The SOCs were 342321413124, and 343231424121 respectively. One SOC was used as own; the other SOC was used as other. This was counterbalanced across participants. The SOC was made noisy as follows: In a probable trial, the next location was based on a triplet from own. In an improbable trial, the next location was based on a triplet from other. The two probability levels were p = .875 and p = .75, which corresponded to 12 and 24 triplets from other per training blocks of 96 trials. 3.4. Procedure The serial response time (SRT) task consisted of 12 blocks of 98 trials. At the start of each block, two random locations were chosen and the sequence that followed according to the appropriate SOCs was continued for 96 trials. Between blocks participants took a pause of at least ten seconds. Participants were instructed to respond as quickly and accurately as possible by pressing the correct key to the displayed location. When the correct key was pressed, immediately the next location was shown (the response-stimulus interval was 0 ms). If an incorrect key was pressed, a tone was sounded for 100 ms, after which the next location was shown. After the SRT task, the participant performed an inclusion and exclusion task. At the beginning of this phase, the participant was reminded that with good performance 20 euro could be earned. Three performance criteria were described: (1) do not generate repetitions, (2) generate a varied sequence, and (3) generate a sequence that resembles (or does not resemble, viz. in the exclusion task) the sequence shown earlier. Payment was based on the following criteria: maximally three rep-
2 Following Fu et al. and Wilkinson and Shanks (2004), participants were excluded from the analysis if they generated a repeating pattern (such as 12341234 . . . ) in the inclusion or exclusion task. Seven participants produced a repeating pattern.
630
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
etitions and 55% correct, averaged over the inclusion and exclusion tasks. During the test phase, first two random locations were shown, and the participant pressed the corresponding keys. Then a sequence of 96 locations was generated by the participant. The order of administration of the inclusion and exclusion tasks was counterbalanced across participants. Finally, a recognition task was administered. The recognition task followed the procedure of Destrebecqz and Cleeremans (2001). First, the participant performed the SRT task for a single triplet. Then the participants rated, on a scale of 1–6, how likely this triplet was part of the learning sequence. The participants rated all 36 possible triplets, administered in a random order. 4. Results The number of participants in each condition was: n = 18 in one-reversal, p = .75; n = 18 in one-reversal, p = .875; n = 16 in four-reversal, p = .75; n = 19 in four-reversal, p = .875. 4.1. SRT Task In the SRT task, 0.82% of the trials had RTs greater than 1000 ms; these trials were dropped from the analyses. Fig. 1 shows the mean RTs to correct trials from own and other over blocks for the p = .75 and p = .875 conditions (collapsed over the oneand four-reversal conditions as these were not significantly different, see results below). An ANOVA3 was performed on RT with triplet (own vs. other) and blocks (1–12) as within-subjects factors, and noise (p = .75 vs. p = .875) and SOC_type (onereversal vs. four-reversal) as between-subjects factors. The results were as follows: There were no main effects nor interactions of SOC_type, so reversal frequency did not affect RT to own and other. There was a main effect of triplet, F(1, 67) = 115, p < .001, indicating that responses to own triplets were faster than responses to other triplets. There was a main effect of block, F(1, 737) = 74.6, p < .001, so RTs decreased over blocks. There was a triplet noise interaction, F(1, 67) = 8.18, p = .006. This interaction indicated that the difference between RTs to own and other was larger in the p = .875 conditions than in the p = .75 conditions. Finally, there was a triplet block interaction, F(1, 737) = 9.39, p < .001, indicating that RTs to own decreased more over blocks than did RTs to other. These results replicate standard findings in sequence learning research. Next we analyze the RTs to reversal and non-reversal trials in the SRT task to establish whether reversal frequency affects RTs during sequence learning. Fig. 2 shows the mean RT for reversals and non-reversals over blocks for the one-reversal and four-reversal conditions (collapsed over the noise conditions because there were no significant differences between these conditions, see results below). An ANOVA on RT with reversal and block as within-subject factors, and SOC_type and noise as between-subjects factors showed these results: there was neither a main effect nor an interaction of noise with any of the other factors. There was a significant reversal SOC_type interaction, F(1, 67) = 10.63, p = .002, indicating a different pattern of RTs to reversals and non-reversals in the one- and four-reversal conditions. A separate ANOVA in the one-reversal conditions revealed no significant main effect of reversal, F(1, 34) = 2.55, p = .120. However, a similar ANOVA in the four-reversal conditions, indicated faster responses to reversals than to non-reversals F(1, 33) = 10.4, p = .003. Fig. 2 also suggests that early in the one-reversal conditions, responses were slower for reversals than for non-reversals. An ANOVA on the first block of the one-reversal conditions, with reversal as within-subjects factor and noise as between-subjects factor, confirms this by showing a main effect of reversal, F(1, 34) = 10.24, p = .003. 4.2. Inclusion exclusion task To study the extent of explicit knowledge, below we analyze the proportions of own and other that were generated in the inclusion and exclusion task. The sequences generated by the participants were classified as triplets from own, other, neither or invalid.4 Fig. 3 shows the proportions of own, other and neither generated in the inclusion and exclusion task for the one- and four-reversal conditions separately (collapsed over noise levels as we found no significant effects of that factor, see analyses below). On average, 5.40 (SE = .53) triplets were classified as invalid, because these contained repetitions. An ANOVA on the number of invalid triplets, with instruction (inclusion vs. exclusion) as within-subjects factor, and noise (p = .75 vs. p = .875) and SOC_type (one- vs. four-reversal) as between-subjects factors, revealed no significant effects. This indicated that participants generated similar numbers of valid and invalid triplets across conditions. In the succeeding analyses, triplet proportions are relative to the total number of own, other, and neither generated by a participant in a task. In the analyses below, chance level is represented as an equal number of own, other and neither (each having a proportion of one third). An ANOVA on the proportions of generated triplets, with triplet (own vs. other) and task (inclusion vs. exclusion) as within-subject factors, and noise and SOC_type as between-subjects factors, revealed the following: there was no main effect of triplet, F(1, 67) = .65, p = .421, so the proportion of own was not different from the proportion of other. There was a main effect of SOC_type, F(1, 67) = 39.45, p < .001, indicating that more own and other were generated in the one-reversal conditions than in the four-reversal conditions. Follow-up ANOVAs indicated that both in the one-reversal conditions, F(1, 34) = 149.01,
3
If any within-subjects factor or combination of factors violated Mauchly’s Test of sphericity, then for the relevant F-tests lower bound significance was used. Fu et al. calculated neither by subtracting the sum of own and other from the total number of generated triplets (which was 96). This method could classify invalid triplets (such as triplets containing repetitions) as neither. 4
631
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
550 400
400
450
450
500
500
550
own other
2
4
6
8
10
12
2
4
6
8
10
12
Fig. 1. Mean RTs for own and other across blocks in the SRT task. Error bars depict standard errors. Left p = .75; right p = .875.
550 500 450 400
400
450
500
550
reversals non-reversals
2
4
6
8
10
12
2
4
6
8
10
12
0.4 0.3 0.2 0.1 0.0
0.0
0.1
0.2
0.3
0.4
Fig. 2. Mean RTs for reversals and non-reversals across blocks in the SRT task. Error bars depict standard errors. Left one-reversal SOC; right four-reversal SOC.
inclusion task
inclusion task
exclusion task own
neither
other
chance level
exclusion task
Fig. 3. Proportion of own, other, and neither generated in the inclusion exclusion task. Error bars depict standard errors. Left one-reversal SOC; right fourreversal SOC.
632
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
p < .001, and in the four-reversal conditions, F(1, 33) = 5.44, p = .026, generation of own and other exceeded chance levels. There was a main effect of task, F(1, 67) = 5.61, p = .021, indicating that fewer own and other triplets were generated in exclusion than in inclusion. There was no SOC_type task interaction, F(1, 67) = .722, p = .40, so the decrease in own and other was similar in the one- and four-reversal conditions. There was no noise task interaction, F(1, 67) = .72, p = .40, so the decrease in own and other was similar at both noise levels. There was a marginally significant task triplet interaction F(1, 67) = 3.50, p = .066, indicating that the decrease from inclusion to exclusion task was slightly larger for other than for own. To test whether the reversal frequencies of the SOCs were replicated in the generation task and to test for explicit knowledge of reversals, we performed a separate analysis on the proportion of reversals that were generated. An ANOVA was performed on the proportion of reversals, with task as within-subjects factor, and noise and SOC_type as between-subjects factors. There was a main effect of SOC_type, F(1, 67) = 20.484, p < .001, indicating that a smaller proportion of reversals was generated in the one-reversal conditions (M = .21, SE = .02) than in the four-reversal conditions (M = .30, SE = .02). Follow-up analyses in the one- and four-reversal conditions separately, revealed the following: in the one-reversal conditions, generation of reversals was below chance level, F(1, 34) = 168.10, p < .001. In the one-reversal conditions, there was a marginally significant effect of task, F(1, 34) = 3.809, p = .059, indicating an increase in the proportion of reversals from inclusion to exclusion task. In the four-reversal conditions however, generation of reversals was only marginally smaller than chance levels, F(1, 33) = 3.27, p = .08. With four reversals, there was no main effect of task, indicating that the proportion of reversals did not change from inclusion to exclusion task, F(1, 33) = .001, p = .98. 4.3. Recognition task Due to a software error, data of the recognition task was only available for 40 participants (of a total of 71), so the results below are restricted to these participants. On the response times to the presented triplets, an ANOVA was performed, with triplet (own, other or neither), reversal and location (first, second, or third location) as within-subject factors, and SOC_type and noise as between-subjects factors. There was a main effect of Noise, F(1, 36) = 8.35, p = .006, indicating overall faster responses in the high-noise conditions than in the low-noise conditions. Furthermore, there was a main effect of location, F(1, 72) = 26.26, p < .001. Separate analyses revealed faster responses on the second location than on the first, F(1, 36) = 14.52, p = .001, but no difference in RTs between the second and third location F(1, 36) = .23, p = .64. There was a reversal SOC_type interaction F(1, 36) = 4.66, p = .038. Separate analyses for the one- and four-reversal conditions revealed that in the one-reversal conditions response times did not differ between reversals and non-reversals, F(1, 20) = .34, p = .56, whereas they were faster for reversals than non-reversals in the four-reversal conditions, F(1, 16) = 7.59, p = .014. After the reaction time phase of a recognition trial, the presented triplet was rated on a Likert scale. Fig. 4 shows the recognition ratings for own, other, and neither. After recoding, the scores in Fig. 4 run from 3.5 (the triplet certainly did not belong to the training material) to 3.5 (the triplet certainly belonged to the training material). An ANOVA on the recognition ratings with triplet (own, other, or neither) and reversal (reversal, non-reversal) as within-subjects factors, and SOC_type and noise as between-subject factors resulted in a triplet reversal SOC_type interaction, F(2, 72) = 5.97, p = .004. This interaction was further explored through separate analyses for the one- and four-reversal conditions.
1.0 0.5 0.0 −0.5 −1.0
−1.0
−0.5
0.0
0.5
1.0
reversals non-reversals
1.5
1.5
4.3.1. One-reversal conditions In the one-reversal conditions, there was a marginal main effect of reversal, F(1, 20) = 3.52, p = .076, indicating that reversals were rated as more familiar than non-reversals, and, more interestingly, a triplet reversal interaction F(2, 40) = 4.80, p = .014. Inspection of Fig. 4 indicates that this interaction is due to the difference in ratings between the neither reversal and
own
other
neither
own
other
neither
Fig. 4. Recognition ratings to reversals and non-reversals of own, other, and neither. Error bars depict standard errors. Left one-reversal SOC; right fourreversal SOC.
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
633
non-reversal triplets. Separate t-tests confirm this by showing no significant differences between any pair of data points (p’s > .1), except for the t-test between neither reversals and non-reversals, t(21) = 4.00, p = .001. Additionally, separate one-sample t-tests revealed that all data points differed from 0 (the middle of the scale), with the neither non-reversals rated below the middle, and the others rated above the middle of the scale (p’s < 0.01). To examine whether similar effects would be present in the inclusion exclusion task, we performed an ANOVA on generation of neither, with reversal and task as within-subject factors, and noise and SOC_type as between-subject factors. There was a SOC_type reversal task interaction, F(1, 67) = 7.14, p = .009. Separate analyses for the one-and four-reversal conditions revealed, that in the one-reversal conditions there was a reversal task interaction, F(1, 34) = 9.43, p = .004, which was non-significant in the four-reversal conditions, F(1, 33) = .655, p = .424. Separate analyses for reversals and non-reversals in the one-reversal condition indicated that there was an increase in the proportion of neither reversals, F(1, 34) = 9.20, p = .005, but no change in the proportion of neither non-reversals F(1, 34) = 1.37, p = .251. 4.3.2. Four-reversal conditions Continuing with the analysis of the recognition ratings; in the four-reversal conditions there was a main effect of reversal, F(1, 16) = p = .043, indicating that reversals were rated as more familiar than non-reversals, and there was a main effect of triplet, F(2, 32) = 3.34, p = .048. Separate t-tests revealed that own, t(17) = 2.55, p = .021, and other, t(17) = 2.56, p = .020, were rated above the middle of the scale, but ratings of neither, t(1, 17) = 1.70, p = .107, did not differ from the middle of the scale. Finally, there is a trend apparent that only the reversals of own and other were rated higher than the reversals of neither, while non-reversals were rated similarly. However, this trend should be manifest through a reversal triplet interaction, and this interaction was non-significant, F(1, 32) = 1.87, p = .191. 5. Discussion In this experiment we examined how reversal frequency affects learning of noisy SOCs, both in performance measures (RTs) and in subsequent measures of explicit knowledge. Triplets from own were presented with a probability of p = .75 or p = .875, otherwise other was presented. The reversal frequency of own and other was one or four. When own and other contained one reversal, reversal frequency was informative for discriminating own and other from neither. When own and other contained four reversals, reversal frequency was not informative for discriminating own and other from neither. The results of this experiment show a clear distinction between response time measures in the SRT task, generation performance, and recognition ratings. The results from the SRT task were comparable with other sequence learning studies insofar that over blocks of training, an increasing difference between RTs to own versus other was found. This result was found in both the one- and four-reversal conditions, and hence the relatively higher reversal frequency did not prevent sequence learning to take place. In contrast to our expectations, both participants trained on an SOC with one reversal, and participants trained on an SOC with four reversals were able to decrease generation of own and other in exclusion. So, participants showed the ability to control knowledge of the regularities of own and other regardless of reversal frequency. Reversal frequency did affect which knowledge became explicit, and in what way (control or metaknowledge). Early in training, responses to reversals were slower than non-reversals in the one-reversal conditions, but faster in the four-reversal conditions. Over blocks, this difference in RTs remained in the four-reversal conditions, but disappeared in the one-reversal conditions. It appears that, starting early in training, reversal frequency affected RTs to reversals.5 From inclusion to exclusion, participants trained on a one-reversal SOC showed a larger increase of neither reversals than neither non-reversals. So, training on a noisy one-reversal SOC seemed to induce controllable knowledge of neither reversals being absent. The results of the recognition task were not congruent with the results of the inclusion exclusion task. In the one-reversal conditions, only the two non-reversals of neither were rated as less familiar than the other reversals and non-reversals of own, other, and neither. The two non-reversals in the one-reversal neither are both continuous triplets, namely 123 and 321. When trained on a noisy one-reversal SOC, participants showed metaknowledge of these neither continuous being absent, but no metaknowledge besides that. Below we discuss this interesting pattern of results, reflecting on the dissociations between inclusion exclusion task and recognition task in light of the information and sensitivity criterion (Cheesman & Merikle, 1984; Dienes & Berry, 1997). Before doing so, we first examine a curious find, namely that participants did discriminate between own and other in the SRT task, but not in generation or recognition. 5.1. Knowledge of own and other as controlled subjective thresholds Besides the SRT task, two measures of explicit knowledge were administered: the inclusion exclusion task and a recognition task. The current experiment is, in part, a replication of an experiment by Fu et al. (2008) and an experiment 5 Vaquero et al. (2006) observed slower responses to reversals than to non-reversals, both when training on a one-reversal SOCs and a (four-reversal) random sequence. They proposed inhibition of return as a possible explanation. Our results in a noisy SOC experiment with one- and four-reversal SOCs incidate fast learning of reversals when they are likely, and slow learning of reversals when they are unlikely. A similar argument was put forth by Vaquero et al. (2006) with regard to random sequences.
634
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
by Wilkinson and Shanks (2004). Both researchers found a larger number of own than other (I > B). In this experiment we did not replicate this result: generation of own did not exceed generation of other. When using other as baseline, we cannot conclude that any knowledge of own was demonstrated in the inclusion or the exclusion task. The divergence of this finding with previous research is interesting, since the procedures were closely mimicked. However, there were differences in instruction: In this experiment, participants were requested to generate a varied sequence, while in the experiments of Fu et al. and Wilkinson and Shanks, no such instruction was given. If we take a subjective versus objective threshold perspective (Dienes & Berry, 1997), this difference in instruction might be important: without the request for a varied sequence, the subjective threshold for hits vs. false alarms might be so that only the more likely triplets (from own) were generated. The request for a varied sequence might have induced participants to maintain a ‘more lenient’ subjective threshold, accepting both likely (own) and unlikely (other) triplets for generation. If the subjective threshold is lenient enough, then noise levels of p = .875 and p = .75 would result in the same behavior. This was indeed the case. Examining generation of own and other combined, it was found that generation of own and other exceeded chance levels, so participants have demonstrated knowledge of the regularities of own and other. In agreement with Fu et al. (2008), participants acquired control over this knowledge, by generating less own and other in exclusion than in inclusion. In contrast with Fu et al., this occurred both with high (p = .75) and low noise (p = .875) levels. This could be explained by a subjective threshold over which control is exercised; when asked to include, participants generated triplets of which they were less certain that they occurred during training, when asked to exclude, participants generated triplets of which they were more certain that they did not occur during training. In summary, we explain the lack of any difference in generation of own and other as participants applying a lenient subjective threshold, possibly induced by the request for a varied sequence. If this explanation is correct, it could have grave consequences for the noisy SOC paradigm, since it would imply that participants do not experience triplets from other as noise, but as information that can optionally be retrieved. 5.2. About sensitive and informative measures Overall, the pattern of results is intriguing; the different kinds of regularities that could be learned, selectively affected behavior on different measures. In assessing these measures themselves, we invoke the information and sensitivity criteria of Shanks and St. John (1994). The information criterion prescribes that tests of awareness should tap into the exact same knowledge on which performance is based. We have demonstrated that in noisy SOCs, various regularities can be learned that cannot be measured by comparing knowledge of own to knowledge of other. Thus, a noisy SOC paradigm comparing only knowledge of own to knowledge of other might not satisfy the information criterion. The sensitivity criterion, prescribes that tests of awareness should be sensitive to all of the relevant knowledge. We have demonstrated that the inclusion exclusion task and recognition ratings can measure knowledge of different regularities, so these tasks in isolation were not sufficient to measure all of the relevant conscious knowledge. On the contrary, they can measure entirely different knowledge. More possible violations of the sensitivity criterion may be apparent when we look at the effects of low reversal frequency on generation of own and other. Low reversal frequency strongly increased the proportion of generated own and other. Generation is considered (for instance by Jiménez et al., 1996) to be affected both by implicit and explicit knowledge. If the larger generation of own and other in the one-reversal SOCs was primarily based on implicit knowledge, then it is interesting that larger generation was not paired with faster response times for own and other, since it implies that response time was not sensitive to this implicit knowledge. If larger generation of own and other was primarily based on explicit knowledge, then it is interesting that larger generation of own and other in the one-reversal SOCs was not mirrored by better inclusion exclusion performance or higher recognition ratings, since it implies that the latter two were not sensitive to this explicit knowledge.
6. Conclusion We conclude that current methodologies of studying implicit and explicit learning are too restrictive: traditionally, the SOCs of Reed and Johnson (1994) are used to measure knowledge of only one kind of regularity, namely knowledge of own as compared with knowledge of other. The results of Fu et al. (2008) with noisy SOCs already suggested that more knowledge was acquired. The results of this experiment, in which reversal frequency was manipulated independently of SOC structure, indicate that various regularities can be learned and expressed. Learning of these regularities can be regarded as confounds that should be prevented by using more controlled learning materials. Learning of these regularities could also be integrated in current theorizing on implicit learning. The latter standpoint differs from that of Reed and Johnson (1994). Reed and Johnson aimed at developing learning materials that could be used to measure knowledge of only the second order conditional structure of the SOC, and considered various simple event frequencies as a threat to validity. We would emphasize that knowledge of an SOC can be an integration of knowledge of different regularities (a standpoint that is also reflected by connectionist models of learning, such as the simple recurrent network of Cleeremans, Servan-Schreiber, & McClelland, 1989, that learns progressively more complex structures in an associative fashion). Representations of these regularities might interact with one another, affecting the
T. Pronk, I. Visser / Consciousness and Cognition 19 (2010) 627–635
635
extent to which knowledge of a particular regularity can be expressed implicitly or explicitly. By factorially varying such regularities and examining the extent to which knowledge of particular regularities is implicit or explicit, a more nuanced picture can be drawn of the way that sequence knowledge comes to be. Acknowledgments We would like to thank Arnaud Destrebecqz, Juan Lupiáñez, Rianne Hoeks, and an anonymous reviewer for their valuable comments on earlier drafts of this paper. References Cheesman, J., & Merikle, P. M. (1984). Priming with and without awareness. Perception & Psychophysics, 36, 387–395. Cleeremans, A., & McClelland, J. L. (1991). Learning the structure of event sequences. Journal of Experimental Psychology: General, 120, 235–253. Cleeremans, A., Servan-Schreiber, D., & McClelland, J. L. (1989). Finite state automata and simple recurrent networks. Neural Computation, 1(3), 372–381. Cohen, A., Ivry, R. I., & Keele, S. W. (1990). Attention and structure in sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16(1), 17–30. Colman, A. M. (1995). Game theory and its applications in the social and biological sciences. Oxford: Butterworth-Heinemann. Curran, T., Smith, M. D., DiFranco, J. M., & Daggy, A. T. (2001). Structural influences on implicit and explicit sequence learning. In D. L. Medin (Ed.). The psychology of learning and motivation (Vol. 40, pp. 147–182). San Diego, CA: Academic Press. de Houwer, J., Baeyens, F., & Hendrickx, H. (1997). Implicit learning of evaluative associations. Psychologica Belgica, 37, 115–130. Destrebecqz, A., & Cleeremans, A. (2001). Can sequence learning be implicit? New evidence with the process dissociation procedure. Psychonomic Bulletin & Review, 8(2), 343–350. Destrebecqz, A., & Peigneux, P. (2005). Methods for studying unconscious learning. In S. Laureys (Ed.), Progress in brain research (pp. 69–80). Amsterdam: Elsevier. Destrebecqz, A., Peigneux, P., Laureys, S., Degueldre, C., Fiore, G. D., Aerts, J., et al (2005). The neural correlates of implicit and explicit sequence learning: Interacting networks revealed by the process dissociation procedure. Learning and Memory, 12, 480–490. Dienes, Z., & Berry, D. C. (1997). Implicit learning: Below the subjective threshold. Psychonomic Bulletin & Review, 4, 3–23. Frensch, P. A., Buchner, A., & Lin, J. (1994). Implicit learning of unique and ambiguous serial transitions in the presence and absence of a distractor task. Journal of Experimental Psychology: Learning, Memory & Cognition, 20, 567–584. Fu, Q. F., Fu, X., & Dienes, Z. (2008). Implicit sequence learning and conscious awareness. Consciousness and Cognition, 17(1), 185–202. Glymour, C. (2003). Learning, prediction and causal Bayes nets. Trends in Cognitive Sciences, 7, 43–48. Howard, D. V., Howard, J. H., Jr., Japikse, K. C., DiYani, C., Thompson, A., & Somberg, R. (2004). Implicit sequence learning: Effects of level of structure, adult age, and extended practice. Psychology and Aging, 12, 634–656. Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Journal of Memory and Language, 30(5), 513–541. Jiménez, L., Méndez, C., & Cleeremans. (1996). Comparing direct and indirect measures of sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22(4), 948–969. Jiménez, L., Vaquero, J. M. M., & Lupiáñez, J. (2006). Qualitative differences between implicit and explicit sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32(3), 475–490. Kelly, S. W., Burton, A. M., Riedel, B., & Lynch, E. (2003). Sequence learning by action and observation: Evidence for separate mechanisms. British Journal of Psychology, 94, 355–372. Koch, I., Philipp, A. M., & Gade, M. (2006). Chunking in task sequences modulates task inhibition. Psychological Science, 17(4), 346–350. McShane, J. (1991). Cognitive development: An information processing approach. Oxford: Blackwell. Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of learning: Evidence from performance measures. Cognitive Psychology, 19, 1–32. Perruchet, P., & Amorim, M. A. (1992). Conscious knowledge and changes in performance in sequence learning: Evidence against dissociation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 785–800. Pinker, S. (1994). The language instinct. New York: Morrow. Rauch, S. L., Savage, C. R., Brown, H. D., Curran, T., Alpert, N. M., Kendrick, A., Fischman, A. J., & Kosslyn, S. M. (1995). A PET investigation of implicit and explicit sequence learning. Human Brain Mapping, 3, 271–286. Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Experimental Psychology: General, 118(3), 219–235. Reed, J., & Johnson, P. (1994). Assessing implicit learning with indirect tests: Determining what is learned about sequence structure. Journal of Experimental Psychology: Learning, Memory and Cognition, 20(3), 585–594. Roediger, H. L. III, (1990). Implicit memory: Retention without remembering. American Psychologist, 45, 1043–1056. Shanks, D. R., Holyoak, K. J., & Medin, D. L. (Eds.). (1996). Causal learning. San Diego: Academic Press. Schvaneveldt, R. W., & Gomez, R. L. (1998). Attention and probabilistic sequence learning. Psychological Research, 61(3), 175–190. Shanks, D. R., & Johnstone, T. (1999). Evaluating the relationship between explicit and implicit knowledge in a sequential reaction time task. Journal of Experimental Psychology: Learning, Memory, & Cognition, 25, 1435–1451. Shanks, D. R., & Perruchet, P. (2002). Dissociation between priming and recognition in the expression of sequential knowledge. Psychonomic Bulletin & Review, 9, 362–367. Shanks, D. R., & St. John, M. F. (1994). Characteristics of dissociable human learning systems. Behavioral and Brain Sciences, 17, 367–448. Shanks, D. R., Wilkinson, L., & Channon, S. (2003). Relationship between priming and recognition in deterministic and probabilistic sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 248–261. Stöcker, C., Hoffmann, J., & Sebald, A. (2003). The influence of response-effect compatibility in a serial reaction time task. Quarterly Journal of Experimental Psychology, 56, 685–703. Vandenberghe, M., Schmidt, N., Fery, P., & Cleeremans, A. (2006). Can amnesic patients learn without awareness? New evidence comparing deterministic and probabilistic sequence learning. Neuropsychologia, 44, 1629–1641. Vaquero, J. M. M., Jiménez, L., & Lupiáñez, J. (2006). The problem of reversals in assessing implicit sequence learning with serial reaction time tasks. Experimental Brain Research, 175, 97–109. Werheid, K., Zysset, S., Muller, A., Reuter, M., & von Cramon, D. Y. (2003). Rule learning in a serial reaction time task: an fMRI study on patients with early Parkinson’s disease. Cognitive Brain Research, 16, 273–284. Wilkinson, L., & Shanks, D. R. (2004). Intentional control and implicit sequence learning. Journal of Experimental Psychology: Learning, Memory and Cognition, 30(2), 354–369.
Consciousness and Cognition 19 (2010) 636–643
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Short Communication
Attentional inhibition mediates inattentional blindness Preston P. Thakral *, Scott D. Slotnick Department of Psychology, Boston College, United States
a r t i c l e
i n f o
Article history: Received 12 October 2009 Available online 15 March 2010 Keywords: Spatial attention Visual awareness Task accuracy
a b s t r a c t Salient stimuli presented at unattended locations are not always perceived, a phenomenon termed inattentional blindness. We hypothesized that inattentional blindness may be mediated by attentional inhibition. It has been shown that attentional inhibition effects are maximal near an attended location. If our hypothesis is correct, inattentional blindness effects should similarly be maximal near an attended location. During central fixation, participants viewed rapidly presented colored digits at a peripheral location. An unexpected black circle (the critical stimulus) was concurrently presented. Participants were instructed to maintain central fixation and name each color/digit, requiring focused attention to that location. For each participant, the critical stimulus was presented either near to or far from the attended location (at the same eccentricity). In support of our hypothesis, inattentional blindness effects were maximal near the attended location, but only at intermediate task accuracy. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction A salient stimulus presented at an unattended location is not always perceived, a phenomenon termed inattentional blindness (Mack & Rock, 1998). While the experimental conditions that give rise to inattentional blindness have been explored to some degree, the cognitive processes mediating inattentional blindness are still unknown. Inattentional blindness effects have been directly linked to classic findings in the spatial attention literature. Attentional control settings (Folk, Remington, & Johnston, 1992; Folk, Remington, & Wright, 1994) can affect the rate of inattentional blindness, where, for example, there is greater inattentional blindness for a white stimulus if the task involves attention to black stimuli (Koivisto, Hyona, & Revonsuo, 2004; Koivisto & Revonsuo, 2008; Most, Scholl, Clifford, & Simons, 2005; Most et al., 2001). There is also evidence that inattentional blindness effects reflect selective spatial attention corresponding to maximal facilitation of processing at an attended location (e.g., Downing, 1988; Erikson & St. James, 1986). In a typical inattentional blindness experiment, an observer fixates at the center of the display and simultaneously attends to a demanding perceptual task in the periphery. Surprisingly, participants often fail to perceive a ‘critical stimulus’ presented at fixation, an illustration of inattentional blindness. Mack and Rock (1998) modified this basic paradigm by presenting the critical stimulus at peripheral locations in the display. They found a lower rate of inattentional blindness when the critical stimulus was presented within the attended region versus when it was presented outside of the attended region (see Bridgeman & Lathrop, 2007, for a demonstration of robust inattentional blindness to a large critical stimulus presented outside of the attended region). In an even more detailed study, Newby and Rock (1998) presented the peripheral critical stimulus at varying distances from the locus of attention and found that the rate of inattentional blindness increased parametrically with increasing distance (see also, Most, Simons, Scholl, & Chabris, 2000). These findings are consistent with a gradient model of selective spatial * Corresponding author. Address: Department of Psychology, Boston College, McGuinn Hall, Chestnut Hill, MA 02467, United States. Fax: +1 617 552 0523. E-mail address: [email protected] (P.P. Thakral). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.02.002
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
637
attention, where facilitatory attention effects are maximal at the attended location and gradually fall off with increasing distance. Mack and Rock (1998) predicted that inhibitory processes may play a role in inattentional blindness when they speculated ‘‘anything that causes attention to be inhibited at the location in which the critical stimulus appears predictably increases the likelihood of inattentional blindness” (p. 228). As such, we hypothesized that inattentional blindness may be mediated, in part, by attentional inhibition. Attentional inhibition is modeled by a facilitatory region centered at the attended location (as described above) that is surrounded by a region of inhibition (which is in turn surrounded by an area of little or no attentional modulation). This center-surround model of attention has been supported in a number of behavioral studies (Bahcall & Kowler, 1999; Bichot, Cave, & Pashler, 1999; Caputo & Guerra, 1998; Cepeda, Cave, Bichot, & Kim, 1998; Cutzu & Tsotsos, 2003; Kim & Cave, 1999; Mounts, 2000a; Müller, Mollenhauer, Rösler, & Kleinschmidt, 2005; Müller, Mühlenen, & Geyer, 2007; for a review, see Cave & Bichot, 1999) and neural studies (Boehler, Tsotsos, Schoenfeld, Heinze, & Hopf, 2009; Hopf et al., 2005; Müller & Kleinschmidt, 2004; Slotnick, Hopfinger, Klein, & Sutter, 2002; Slotnick, Schwarzbach, & Yantis, 2003). If our hypothesis that inattentional blindness is mediated by attentional inhibition is correct, the largest inattentional blindness effects for peripheral critical stimuli should be observed near to the attended location. To investigate our hypothesis, we modified a paradigm used by Koivisto et al. (2004) that showed robust inattentional blindness effects. In their paradigm the unexpected stimulus was presented at fixation, while in the present study it was presented in the periphery at the same eccentricity as the attended location (as is typically done in attentional inhibition studies). Specifically, during central fixation participants were rapidly presented with to-be-attended colored digits (which they named) at a peripheral location for two trials. On the third, critical, trial an unexpected black circle, the critical stimulus, was presented along with the colored digits either at central fixation, near to the attended location, or far from the attended location (Fig. 1). It is important to note that only one critical trial occurred for each participant, as the resultant expectation of stimuli at non-target locations is known to foster subsequent perception. The spatial location of the critical stimulus near to the attended location was selected to fall within the expected region of attentional inhibition and the more distant critical stimulus was selected to fall outside the inhibitory region, based on the location of maximal attentional inhibition effects reported in previous paradigms (Bahcall & Kowler, 1999; Mounts, 2000a; Müller et al., 2005).
2. Experiment 1 2.1. Materials and methods 2.1.1. Participants One hundred sixteen undergraduate students at Boston College with normal or corrected-to-normal visual acuity took part in the study. Each participant received either $5 or 1 study pool credit. Informed consent was obtained before the experiment commenced. The experimental protocol was approved by the Boston College Institutional Review Board. 2.1.2. Stimulus and task Our experimental protocol was very similar to that of Koivisto et al. (2004), who observed robust inattentional blindness of unexpected stimuli at fixation. We extended their paradigm to include unexpected stimuli at peripheral locations. Each trial consisted of an initial fixation frame (1500 ms), two selective attention frames (each 350 ms), and a trailing mask frame (500 ms; Fig. 1). For each trial, the black fixation cross (subtending .5° visual angle) was presented continually at the center of a white circular aperture (10.1° visual angle in diameter) until the onset of the mask. During each selective attention frame, to-be-attended colored digits (.5° visual angle in height) were presented at an eccentricity of 3.2° visual angle from fixation, 11.0° polar angle from the horizontal meridian. Digit values (ranging from 1 through 9) and colors (orange, green, blue, yellow, gray, red, pink, purple, and black) were randomly selected for each frame. During all trials, participants were instructed to maintain central fixation and recite each number and its respective color aloud. Responses were recorded for subsequent analysis. Each participant first completed two trials where only the rapidly presented colored digits were presented. These initial trials allowed participants to practice the task and, given that the task was relatively difficult, learn to selectively attend to the location of the digits. Then, on the third and final trial – the critical trial – a black circle (.5° visual angle in diameter) was presented along with the digits (for 700 ms); this served as the critical, unexpected stimulus. For each participant, the circle was presented at one of three locations: (1) 1.5° visual angle from the attended digit location at the same eccentricity from fixation in the ‘near’ condition, (2) 3° visual angle from the attended location at the same eccentricity from fixation in the ‘far’ condition, and (3) at fixation. As mentioned previously, the peripheral critical stimulus locations were selected based on previous work showing maximal attentional inhibition effects 1.5° visual angle from the attended location (Bahcall & Kowler, 1999; Mounts, 2000a; Müller et al., 2005). After these three trials, each participant filled out a questionnaire to determine whether they had been aware of the critical stimulus (Koivisto et al., 2004). The first question asked whether or not anything new had been detected that had not been described during the instructions. The second question asked the participant to choose the correct critical stimulus out of a possible five shapes (circle, triangle, square, diamond, and star). The last question asked the participant to identify
638
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
Fig. 1. Depiction of three critical trial types, each consisting of black fixation cross, to-be-attended colored digits, and critical stimulus (black circle), followed by a mask. (a) Control condition, with critical stimulus presented at fixation. (b) Near condition, with critical stimulus presented 1.5° visual angle from the attended location. (c) Far condition, critical stimulus presented 3° visual angle from the attended location.
the spatial location of the stimulus (by marking the location within a large circle with a fixation cross similar to that shown experimentally). Participants were considered aware of the critical stimulus if they answered ‘‘yes” to the first question and responded accurately to at least one of the last two questions (i.e., the correct shape and proximate location). Two-tailed chisquare tests were used for statistical analysis, except under conditions of sparse data (n 6 5) where Fisher exact tests were employed (Most et al., 2000). 2.1.3. Results Replicating previous findings, robust inattentional blindness effects were seen when the critical stimulus was presented at fixation (7.7% detection, n = 26). Consistent with our hypothesis, the rate of inattentional blindness was greater when the peripheral critical stimulus was near to the attended location (33.3% detection) than when it was far from the attended location (42.2% detection), although this difference was not significant (v2(1) < 1). We reasoned that this non-significant detection rate difference for near versus far peripheral critical stimuli may have been due to variable accuracy across participants, as inattentional blindness effects have been shown to vary as a function of accuracy (Cartwright-Finch & Lavie, 2007; Fougnie & Marois, 2007; Simons & Jensen, 2009; Todd, Fougnie, & Marois, 2005). To investigate this possibility, participants in the peripheral critical stimulus conditions were separated into two accuracy groups using a median split based on digit/color naming accuracy (with a maximum accuracy of 12, computed from the 3 trials 2 digit 2 color responses) such that the number of participants in each group were as similar as possible. The first group had an accuracy of 86.1% (n = 21 for the near condition, n = 18 for the far condition) while the
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
639
Fig. 2. (a–b) Percent detection of critical stimulus near to and far from the attended location as a function of accuracy in Experiment 1. (c–d) Percent detection of critical stimulus near to and far from the attended location, as a function of accuracy in Experiment 2. (e) Percent detection of critical stimulus near to and far from the attended location, as a function of accuracy in Experiment 3.
second group had an accuracy of 59.0% (n = 24 for the near condition, n = 27 for the far condition). As illustrated in Fig. 2a and Fig. 2b, distinct patterns of results were observed for these groups. For the 59.0% accuracy group (Fig. 2a), there was significantly greater inattentional blindness for stimuli near the attended location (16.7% detection) than for stimuli far from the attended location (48.2% detection; p < .05, Fisher exact test). By contrast, for the 86.1% accuracy group (Fig. 2b), percent detection of near and far critical stimuli did not significantly differ (near location 52.4% detection, far location, 33.3% detection; v2(1) = 1.43, p > .20). The interaction between accuracy and location was significant (p < .05, Fisher exact test, computed using number of detections). The accuracy analysis was also conducted using a median split separately for each near and far group (rather than collapsing these groups first) and the identical results were observed. 3. Experiment 2 In Experiment 1, participants with differential accuracy were found to experience different patterns of inattentional blindness. However, it could be argued that our median split procedure did not separate groups by accuracy per se because this factor was not experimentally manipulated. We investigated this possibility in Experiment 2 by varying the interstimulus interval to manipulate accuracy.
640
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
3.1. Materials and methods 3.1.1. Participants One hundred ten undergraduate students at Boston College with normal or corrected-to-normal visual acuity took part in the study (data from six participants were not included in the analysis because they did not perceive the critical stimulus during the full attention trial, defined below). Each participant received either $5 or 1 study pool credit. Informed consent was obtained before the experiment commenced. The experimental protocol was approved by the Boston College Institutional Review Board. 3.1.2. Stimulus and task Unless otherwise stated, the experimental and analysis procedure was identical to Experiment 1. Each participant completed three trials with the critical stimulus (Newby & Rock, 1998), each preceded by two practice trials without the critical stimulus. The first critical trial was identical to that of Experiment 1, where participants had no knowledge that there might be a critical stimulus. This was followed by a divided attention trial where participants were not explicitly instructed to look for the critical stimulus, but the questions asked after the previous critical trial may have alerted them to the possibility that an additional object might appear. In a final full attention trial participants were instructed to actively detect whether anything new appeared on the screen which was expected to result in perception of the critical stimulus. Two interstimulus intervals were selected based on the results of a pilot study with different participants (n = 52), where we varied interstimulus interval and computed the corresponding change in accuracy to match the two accuracy groups of Experiment 1. Unless otherwise specified, the procedure of the pilot study was identical to Experiment 1. To experimentally manipulate accuracy, the method of constant stimuli was used on an individual participant basis to determine the interstimulus interval that yielded the desired accuracy. Specifically, participants completed 36 trials without the critical stimulus, 3 trials at each of 12 interstimulus intervals (16, 48, 80, 120, 150, 250, 350, 450, 550, 650, 750, and 850 ms) presented in random order. Based on each participant’s accuracy as a function of interstimulus interval, linear interpolation was used to compute the interstimulus interval that yielded the target accuracies of 64.9% and 87.1% (these accuracies were taken from the participants, n = 52, who had already completed Experiment 1 when the pilot study commenced). This yielded two specific interstimulus intervals, 248 ms and 487 ms, that were used to experimentally manipulate accuracy in Experiment 2. Specifically, each participant was randomly assigned to one of four possible accuracy groups; 487 ms/near, 487 ms/far, 248 ms/ near, and 248 ms/far (n = 26/group). 3.1.3. Results For critical trials, participants in the 487 ms group had an accuracy of 70.4% and those in the 248 ms group had an accuracy of 55.5%. For the 70.4% accuracy group (Fig. 2c), there was significantly greater inattentional blindness for stimuli near the attended location (46.1% detection) than for stimuli far from the attended location (80.8% detection; p < .05, Fisher exact test). By contrast, for the 55.5% accuracy group (Fig. 2d), the rate of inattentional blindness for near and far critical stimuli did not differ (near location 34.6% detection, far location, 34.6% detection; v2(1) = 1.00, p > .20). The interaction between accuracy and location did not reach significance (v2(1) = 2.82, p = .093, computed using number of detections). As expected, no significant effects were observed in the divided attention trials or full attention trials. 4. Experiment 3 The overall performance in Experiment 2 was lower than that of Experiment 1. Experiment 3 was conducted in an effort to experimentally manipulate accuracy such that it matched the higher accuracy group of Experiment 1. 4.1. Materials and methods 4.1.1. Participants Fifty-two undergraduate students at Boston College with normal or corrected-to-normal visual acuity took part in the study. Each participant received either $5 or 1/2 study pool credit. Informed consent was obtained before the experiment commenced. The experimental protocol was approved by the Boston College Institutional Review Board. 4.1.2. Stimulus and task Unless otherwise stated, the experimental and analysis procedure was identical to Experiment 2. An interstimulus interval of 1310 ms was selected based on the results of the pilot study described in Experiment 2 to match the 86.1% accuracy from Experiment 1 (correcting for the average deviation between predicted accuracies of the pilot study and the observed accuracies of Experiment 2). 4.1.3. Results For critical trials, participants had an accuracy of 87.5% (Fig. 2e), which was comparable to the higher accuracy group of Experiment 1 (86.1%). Similar to the higher accuracy group results of Experiment 1, percent detection of near and far critical
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
641
stimuli did not significantly differ (near location 96.2% detection, far location, 92.3% detection; p > .20, Fisher exact test). No significant effects were observed in the divided attention trials or full attention trials. 5. Discussion In support of the hypothesis under investigation, the differential inattentional blindness results near to versus far from the attended location mirrored those of previous attentional inhibition findings, where maximal inattentional blindness effects were observed when the peripheral critical stimulus was presented near to the attended location. However, as illustrated in Fig. 3, differential inattentional blindness effects were only revealed under conditions of intermediate accuracy (as indicated by a significant quadratic component; v2(1) = 5.40, p < .05). That inattentional blindness effects depend on accuracy can be explained in terms of the spatial extent the attentional window, which can vary depending on task demands (Downing, 1988; Erikson & St. James, 1986; Mounts, 2000b). While previous models of spatial attention have focused on the facilitatory attentional window (Downing, 1988; Erikson & St. James, 1986), the surrounding inhibitory attentional window can vary with task demands as well (Mounts, 2000b). Considered in this framework, participants in Experiment 1 and Experiment 3 who found the task relatively easy, with higher accuracy (86.1% in Experiment 1, 87.5% in Experiment 3), may have had a facilitatory attentional window such that near locations were processed to a greater degree than far locations (although these differences were not significant). Participants who found the task somewhat difficult, with intermediate accuracy (59.0% in Experiment 1 and 70.4% in Experiment 2), may have had a more focused facilitatory attentional window (i.e., at the attended location) with a narrow surrounding inhibitory window such that critical stimuli were perceived less frequently at near versus far locations. When participants found the task very difficult, as with the lowest accuracy group of Experiment 2 (55.5%), the spatial extent of the inhibitory attentional window may have encompassed both near and far locations resulting in robust inattentional blindness at near and far locations. While this interpretation is consistent with the present findings, future work will be needed to assess the degree to which the facilitatory and inhibitory attentional windows are modulated by stimulus and task factors. As mentioned previously, Newby and Rock (1998) demonstrated that the rate of inattentional blindness increased parametrically as distance from the locus of attention increased, which is consistent with a gradient (purely facilitatory) model of selective attention (see also, Most et al., 2000). While, these results might appear inconsistent with the present findings, the presence of an inhibitory component could not be tested in that study as only one critical stimulus was presented outside the attended region. Specifically, there was no ‘far’ location which would have been necessary to reveal effects consistent with attentional inhibition (as with the present near versus far comparisons). As such, the present results have extended these previous findings by assessing inattentional blindness effects further outside the attended region. It is important to consider an important methodological difference between a majority of previous attentional inhibition studies, where the attended location was flanked by task-irrelevant distractors (Bahcall & Kowler, 1999; Bichot et al., 1999; Boehler et al., 2009; Caputo & Guerra, 1998; Cutzu & Tsotsos, 2003; Hopf et al., 2005; Kim & Cave, 1999; Mounts, 2000a; Müller et al., 2005, 2007; Müller & Kleinschmidt, 2004; Slotnick et al., 2002, 2003; for a review, see Cave & Bichot, 1999), and the present study, where the attended stimulus was presented in isolation on the first two trials. It could be argued that the present inattentional blindness effects may not have been mediated by attentional inhibition, given that distractors may be necessary to tag the spatial location of attentional inhibition (i.e., the location of the critical stimulus). However, atten-
Fig. 3. Difference in percent detection of critical stimulus near to versus far from the attended location as a function of accuracy in all three experiments (open circles) and the best-fit second-order polynomial.
642
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
tional inhibition has been shown to occur at spatial locations surrounding the attended location in the absence of distractors. Cepeda et al. (1998) reported significantly slower reaction times to probes presented at non-distractor locations (1.5° of visual angle from the attended location), and attentional inhibition of spatial locations without distractors has also been observed using fMRI (Sylvester, Jack, Corbetta, & Shulman, 2008). This evidence indicates that attentional inhibition can occur at spatial locations with no distractors, thus, the current inattentional blindness results can be attributed to attentional inhibition. Although not related to the hypothesis under investigation, we found the greatest degree of inattentional blindness when the critical stimulus was presented at fixation (7.7% detection in Experiment 1), replicating previous results that employed similar paradigms (Koivisto et al., 2004; Mack & Rock, 1998). It is important to note that when the target was presented at fixation it entailed an object changing shape (fixation cross to circle), whereas in the peripheral conditions, the critical stimulus had an abrupt onset, the latter of which is known to have relatively higher attentional salience (e.g., Yantis & Hillstrom, 1994; Yantis & Jonides, 1984). This would be expected to produce relatively higher rates of detection for more peripheral onset targets. In addition, rates of detection at fixation may have been lower as this location was the farthest away from the attended location. Recent findings have also demonstrated greater task-irrelevant suppression at central versus peripheral locations (Chen & Treisman, 2008). Even considering these factors, it is striking that detection rates at fixation were so low which supports Mack and Rock (1998) who stated ‘‘there is no conscious perception without attention” (p. 227). It could be argued that the low rate of detection near to the attended location in Experiments 1 and 2 (for the intermediate accuracy conditions) did not reflect attentional inhibition, but rather was due to a generally lower degree of attentional engagement during the task (which would also predict low accuracy and a low rate of detection). However, a general failure to attend would predict similar rates of detection at the near and far locations. Our differential rates of detection at near and far locations (Fig. 2a and c) can be taken as evidence against an explanation based on general inattention (a similar argument can be made against other general factors, such as poor visual acuity or color blindness). It is also important to consider eyemovements as a possible explanation of these effects. Due to cortical magnification (see, Slotnick, Klein, Carney, & Sutter, 2001), if participants did make an eye-movement to the attended location, processing of stimuli at that location would be amplified the most, processing of the near target location would be amplified to a lower degree, and processing of the far target location would be amplified to an even lower degree. As such, the eye-movement account would predict the highest rate of target detection near to the attended location. Exactly the opposite was observed, where the nearest location had the lowest rate of detection (under conditions of intermediate accuracy, Fig. 2a and c) thus arguing against the possibility that the present results were due to eye-movements. Our finding that attentional inhibition mediates inattentional blindness under conditions of intermediate accuracy has implications for present theories of spatial attention. Specifically, the present inattentional blindness evidence suggests that a complete model of spatial attention should include both facilitatory and inhibitory components (i.e., a center-surround model) and that the spatial extent and magnitude of these components are task dependant. By manipulating task factors (i.e., systematically varying accuracy) our results have provided some evidence detailing the conditions under which inhibitory effects of attention become appreciable in inattentional blindness paradigms. Acknowledgments We would like to thank Hiram Brownell for statistical advice and Manu Thakral for assisting with stimulus development. References Bahcall, D. O., & Kowler, E. (1999). Attentional interference at small spatial separations. Vision Research, 39, 71–86. Bichot, N. P., Cave, K. R., & Pashler, H. (1999). Visual selection mediated by location: Feature-based selection of noncontiguous locations. Perception & Psychophysics, 61, 403–423. Boehler, C. N., Tsotsos, J. K., Schoenfeld, M. A., Heinze, H.-J., & Hopf, J. M. (2009). The center-surround profile of the focus of attention arises form recurrent processing in visual cortex. Cerebral Cortex, 19, 982–991. Bridgeman, B., & Lathrop, B. (2007). Interactions between cognitive space and motor activity. In F. Mast & L. Jänke (Eds.), Spatial processing in navigation, imagery and perception (pp. 107–117). Berlin: Springer. Caputo, G., & Guerra, S. (1998). Attentional selection by distractor suppression. Vision Research, 38, 669–689. Cartwright-Finch, U., & Lavie, N. (2007). The role of perceptual load in inattentional blindness. Cognition, 102, 321–340. Cave, K. R., & Bichot, N. P. (1999). Visuospatial attention: Beyond a spotlight model. Psychonomic, Bulletin, & Review, 6, 204–223. Cepeda, N. J., Cave, K. R., Bichot, N. P., & Kim, M. (1998). Spatial selection via feature-driven inhibition of distractor lesions. Perception & Psychophysics, 60, 727–746. Chen, Z., & Treisman, A. (2008). Distractor inhibition is more effective at a central that at a peripheral location. Perception and Psychophysics, 6, 1081–1091. Cutzu, F., & Tsotsos, J. K. (2003). The selective tuning model of attention: Psychophysical evidence for a suppressive annulus around an attended item. Vision Research, 43, 205–219. Downing, C. J. (1988). Expectancy and visual-spatial attention: Effects on perceptual quality. Journal of Experimental Psychology: Human Perception and Performance, 14, 188–202. Erikson, C. W., & St. James, J. D. (1986). Visual attention within and around the field of focal attention: A zoom lens model. Perception & Psychophysics, 40, 225–240. Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntary covert orienting is contingent on attentional control settings. Journal of Experimental Psychology: Human Perception and Performance, 18, 1030–1044. Folk, C. L., Remington, R. W., & Wright, J. H. (1994). The structure of attentional control: Contingent attentional capture by apparent motion, abrupt onset, and color. Journal of Experimental Psychology: Human Perception and Performance, 20, 317–329. Fougnie, D., & Marois, R. (2007). Executive working memory load induced inattentional blindness. Psychonomic Bulletin & Review, 17, 142–147.
P.P. Thakral, S.D. Slotnick / Consciousness and Cognition 19 (2010) 636–643
643
Hopf, J. M., Boehler, C. N., Luck, S. J., Tsotsos, J. K., Heinze, S. J., & Schoenfeld, M. A. (2005). Direct neurophysiological evidence for spatial suppression surrounding the focus of attention in vision. Proceedings of the National Academy of Sciences of the United States of America, 103, 1053–1058. Kim, M.-S., & Cave, K. R. (1999). Top-down and bottom-up attentional control: On the nature of interference from a salient distractor. Perception & Psychophysics, 61, 1009–1023. Koivisto, M., Hyona, J., & Revonsuo, S. (2004). The effects of eye movements, spatial attention, and stimulus features on inattentional blindness. Vision Research, 44, 3211–3221. Koivisto, M., & Revonsuo, S. (2008). The role of unattended distractors in sustained inattentional blindness. Psychological Research Psychologische Forschung, 72, 39–48. Mack, A., & Rock, I. (1998). Inattentional blindness. Cambridge, MA: MIT Press. Most, S. B., Scholl, B. J., Clifford, E. R., & Simons, D. J. (2005). What your see is what you set: Sustained inattentional blindness and the capture of awareness. Psychological Review, 112, 217–242. Most, S. B., Simons, D. J., Scholl, B. J., & Chabris, C. F. (2000) Sustained inattentional blindness: The role of location in the detection of unexpected dynamic events. Psyche, 6, Article 14. . Most, S. B., Simons, D. J., Scholl, B. J., Jimenez, R., Clifford, E., & Chabris, C. F. (2001). How not to be seen: The contribution of similarity and selective ignoring to sustained inattentional blindness. Psychological Science, 12, 9–17. Mounts, J. R. (2000a). Evidence for suppressive mechanisms in attentional selection: Feature singletons produce inhibitory surrounds. Perception & Psychophysics, 62, 969–983. Mounts, J. R. (2000b). Attentional capture by abrupt onsets and feature singletons produces inhibitory surrounds. Perception & Psychophysics, 62, 1485–1493. Müller, N. G., & Kleinschmidt, A. (2004). The attentional ‘spotlight’s’ penumbra: Center surround modulation in striate cortex. NeuroReport, 15, 977–980. Müller, N. G., Mollenhauer, M., Rösler, A., & Kleinschmidt, A. (2005). The attentional field has a Mexican hat distribution. Vision Research, 45, 1129–1137. Müller, H. J., Mühlenen, A. V., & Geyer, T. (2007). Top-down inhibition of search distractors in parallel visual search. Perception & Psychophysics, 69, 1373–1388. Newby, E., & Rock, I. (1998). Inattentional blindness as a function of proximity to the focus of attention. Perception, 27, 1025–1040. Simons, D. J., & Jensen, M. S. (2009). The effects of individual differences and task difficulty on inattentional blindness. Psychonomic Bulletin & Review, 16, 398–403. Slotnick, S. D., Hopfinger, J. B., Klein, S. A., & Sutter, E. E. (2002). Darkness beyond the light: Attentional inhibition surrounding the classic spotlight. NeuroReport, 13, 773–778. Slotnick, S. D., Klein, S. A., Carney, T., & Sutter, E. E. (2001). Electrophysiological estimate of human cortical magnification. Clinical Neurophysiology, 112, 1349–1356. Slotnick, S. D., Schwarzbach, J., & Yantis, S. (2003). Attentional inhibition of visual processing in human striate and extrastriate cortex. NeuroImage, 19, 1602–1611. Sylvester, C. M., Jack, A. I., Corbetta, M., & Shulman, G. L. (2008). Anticipatory suppression of nonattended locations in visual cortex marks target location and predicts perception. Journal of Neuroscience, 28, 6549–6556. Todd, J. J., Fougnie, D., & Marois, R. (2005). Visual short-term memory load suppresses temporo-parietal junction activity and induces inattentional blindness. Psychological Science, 16, 965–972. Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture: Evidence from equiluminant visual objects. Journal of Experimental Psychology: Human Perception and Performance, 20, 95–107. Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective attention: Evidence from visual search. Journal of Experimental Psychology. Human Perception and Performance, 10, 601–621.
Consciousness and Cognition 19 (2010) 644–652
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Short Communication
Attribute preference and selection in multi-attribute decision making: Implications for unconscious and conscious thought Narayanan Srinivasan *, Sumitava Mukherjee Centre of Behavioral and Cognitive Sciences, University of Allahabad, India
a r t i c l e
i n f o
Article history: Received 17 November 2009 Available online 27 March 2010 Keywords: Decision making Unconscious thought Sub-sampling Deliberation-without-attention WADD
a b s t r a c t Unconscious thought theory (UTT) states that all information is taken into account and the attributes are weighted optimally resulting in better decisions in complex decision problems during unconscious thought. Very few studies have investigated the actual amount of information processed in the unconscious thought condition. We hypothesized that only a small subset of information might be considered during unconscious thought (like conscious thought). To test this possibility and to explore the way attribute information is selected and combined, we performed computer simulations on the datasets used by previous researchers. The simulations showed that considering a small subset (3–4) of attributes, yields results comparable to previous studies. There is no need to posit infinite capacity in the unconscious thought condition. The results also suggest that weight information is used for attribute selection that could potentially explain the difficulties in replicating the deliberation-without-attention effect. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction
Traditionally the benefits of conscious deliberation in making a complex decision have been emphasized over quick unconscious decision making. However, recent studies based on unconscious thought theory (UTT) have suggested that decision makers should leave complex decisions to the unconscious for making better decisions (Dijksterhuis, 2004; Dijksterhuis, Bos, Nordgren, & van Baaren, 2006; Dijksterhuis, Bos, van der Leij, & van Baaren, 2009; Dijksterhuis & Nordgren, 2006). The UTT (Dijksterhuis & Nordgren, 2006) defines unconscious thought as object-relevant or task-relevant cognitive or affective thought processes that occur while attention is directed elsewhere. Dijksterhuis et al. (2006) presented information about four objects described by four (simple decision problem) or 12 attributes (complex decision problem) and asked participants to make the ‘best’ choice followed by either a period of conscious deliberation or a period of unconscious thought. More participants made the best decision in the unconscious thought condition with complex problems (deliberation-without-attention effect). The UTT argues that is due to the optimal weighting of attributes during high capacity unconscious thought, where all the attributes are considered for making a decision in contrast to the limited capacity conscious thought that focuses only on a subset of available information (Dijksterhuis & van Olden, 2006). The hypothesis that unconscious thought does indeed consider almost all of the information, has not been directly tested till date and has been inferred solely based on better performance in the unconscious thought conditions. Alternatively, it could indeed be possible that unconscious thought (also) focuses on a subset of available information (attributes) while deciding on a complex decision problem. A related suggestion has been made in the attention literature by Myczek and Simons (2008) who suggested that people might employ sub-sampling to judge statistical properties of objects in a display. * Corresponding author. Fax: +91 5322460738. E-mail address: [email protected] (N. Srinivasan). 1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.002
645
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
Through simulations, Myczek and Simons (2008) argued that the existing evidence for mean judgments (Chong & Treisman, 2005) could be explained by focused attention strategies in which people focus on a small subset of the objects to estimate the mean size. Using this as an analogy, one could argue that sub-sampling could occur in the unconscious thought condition as well. In a similar vein, Rey, Golstein, and Perruchet (2009) has argued that only a small number of attributes might be considered in unconscious thought. They have argued that the amount of time available for deliberation is critical and might affect the number of attributes used in making a decision. Lower deliberation and attentional allocation (low or high) may determine the attribute selection processes leading to fewer attributes being considered in the unconscious thought condition compared to the conscious thought condition. In addition to the number of attributes, it is also not clear how information about attributes is combined in arriving at a decision unconsciously. One possible strategy is the weighted additive decision strategy (WADD) in which every attribute is attached a weight (level of importance) and then the weights of the attributes for each choice option is summed up (Newell, Wong, Cheung, & Rakow, 2008). The option that has the largest weight is chosen as the ‘best’. Dijksterhuis and Nordgren (2006) have suggested that WADD is too complex to be performed during unconscious thought and actually used TALLY (best choice is the one with the highest number of positives) to find the best alternative (Dijksterhuis et al., 2006). It should also be noted that efforts to replicate the deliberation-without-attention-effect have failed to find an advantage for unconscious decision making (Acker, 2008; Calvillo & Penaloza, 2009; Newell et al., 2008; Thorsteinson & Withrow, 2009). In the context of the discrepancies in the empirical findings on unconscious decision making and to determine the decision strategies used in unconscious thought, we decided to perform computer simulations to evaluate the possibility of subsampling of attributes in making a complex decision. We propose that the results from the empirical studies on unconscious decision making (Acker, 2008; Dijksterhuis et al., 2006; Newell et al., 2008) does not necessitate that almost all attributes be considered for arriving at the correct choice (which also is a prime assumption in UTT). We hypothesized that considering a subset of attributes might suffice for arriving at the best choice most (around 70–80%) of the time and we wanted to know how much information (number of attributes) is needed for such a performance. This is similar to the suggestions made by Rey et al. (2009) and the simulations would also provide information about the selection of attributes, the number of attributes used to make decisions and the way information about attributes are combined. To test this hypothesis, we performed computer simulations on the datasets used previously in Experiment 1 of Dijksterhuis et al. (2006), Experiments 1 and 3 of Newell et al. (2008) and Rey et al. (2009). If the simulations produce comparable results to those studies with the selection of a subset of attributes, then it is possible for people to arrive at an optimal decision without considering all of the attributes in a multi-attribute decision problem following unconscious thought. If consideration of all attributes were needed to make an optimal decision, then the simulations would produce results comparable to empirical results only with the selection of almost all of the attributes. The simulations were also expected to shed more light on the decision strategies used in unconscious thought (Dijksterhuis & Nordgren, 2006). 2. Method The simulations were performed using R (R Development Core Team, 2008). The data sets (see Appendix A for information) are structured in the following manner: ‘‘Name of Attribute”, weight, Attribute-value for A, Attribute-value for B, Attribute-value for C, Attribute-value for D, where A–D are the choice options. The weights of the attributes were obtained separately from a different pool of participants. Given that the attributes are weighted, the weights could guide the initial choice of which attributes are to be considered (probability-based sampling) and at the stage in which the attribute information is combined to make a decision (WADD). In probability-based sampling, we calculated the probability of an attribute being chosen based on the weights (see Tables 1 and 2). Probability of the attributes was computed by using a simple transformation [1/(Highest_value_of_scale–Weight_of_attribute)] and then normalized to range from 0 to 1 (Highest_value_of_scale = 20 in dataset1 and 10 in Dataset 2). We computed the frequency of choice (percentage of trials in which a particular alternative is chosen) in the simulations. Table 1 Weights of attributes and the respective probability values of the attributes that are used to choose the attributes during probability-based sampling of Dataset 1. Attribute
Weight
NW = 1/(20-weight)
Prob = NW/Sum(NW)
Gas mileage Handling Environment friendly Sound system Service Ease of shifting gears Trunk space Legroom New Available in different colors Has sunroof Has cupholders
18.3 16.5 15.6 14.6 14.3 12.9 12.3 11.8 10.2 6.1 5.9 1.6
0.58824 0.28571 0.22727 0.18159 0.17544 0.14085 0.12987 0.12195 0.10204 0.07194 0.07092 0.05435
.274 .133 .106 .086 .082 .066 .60 .057 .047 .033 .033 .025 P Prob = 1.0
646
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
Table 2 Weights of attributes and the respective probability value of the attributes that are used to choose the attributes during probability-based sampling of Dataset 2. Attribute
Weight
NW = 1/(10-weight)
Prob = NW/Sum(NW)
Security of building Rent Crime rate of area Flatmate is friend Size of apartment Kindness of neighbors View Built-in wardrobe Direction Leisure facilities
8.95 8.60 8.36 7.91 7.56 5.41 5.18 4.70 4.61 4.59
0.9524 0.7143 0.6098 0.4785 0.4098 0.2179 0.2075 0.1887 0.1855 0.1848
.2295 .1722 .1470 .1153 .0988 .0525 .0500 .0455 .0447 .0445 R Prob = 1.0
In probability-based sampling, weights were used to choose the attributes (after the transformation and normalization) and in random sampling, weights were not used in the sampling process. The rationale for using probability-based sampling is to closely simulate variability in selecting the attributes to be considered in the decision process because of individual differences and bounded rationality. Empirical research has shown that people are not strictly rational (Newell, 2005) and hence one may not always choose the attributes strictly according to its (mean) weight. This idea is captured when attributes are sampled based on a probability that is derived from the mean weight of the attribute. This means a higher weighted attribute has more chances of getting picked than a lower rated attribute but the higher weighted attribute does not get picked always before a lower rated one. The programs chose n (1 to maximum) attributes sequentially using two methods: probability-based sampling and random sampling. We used both TALLY (where weights were not considered) and WADD (where weights were considered) with the attributes chosen using probability-based sampling as well as random sampling resulting in four different simulations: (a) probability-based sampling followed by WADD, (b) probability-based sampling followed by TALLY, (c) random sampling followed by WADD, and (d) random sampling followed by TALLY. When certain subsets of attributes were considered, ties occurred between the choice options and in such scenarios; the corresponding simulated trial was rejected. For each n (number of attributes ranging from 1 to maximum), 1500 trials were simulated and we computed the percentage of times a specific choice was made across the trials as a function of the number of attributes. 2.1. Simulation 1 The first simulation was performed on the dataset used by Dijksterhuis et al. (2006), Newell et al. (2008) and Rey et al. (2009). The dataset (see Appendix A, Dataset 1) comprised of four hypothetical Japanese cars: Hatsdun, Kaiwa, Dasuka and Nabusi with each defined by twelve attributes (like ‘Has cupholders’). Weights of the attributes were not considered in Dijksterhuis et al. (2006) and Newell et al. (2008) (TALLY was used to predict the best car) but Rey et al. (2009) obtained the weights for the attributes that has been used in the current simulations. The attribute value for a given car is 1 (attribute is present) and 1 (attribute is absent). For example, [‘‘New”, 10.2, 1, 1, 1, 1] means that ‘‘New” is an attribute with a weight of 10.2 (on a scale of 20) and Hatsdun is new, Kaiwa is not new, Dasuka is new and Nabusi is not new. The program choose n (=1–12) attributes and the percentage of times each choice was made for different number of attributes was calculated. The probabilities of the attributes (dataset 1) that were used in the simulations (for probability-based sampling) is shown in Table 1. 2.2. Simulation 2 The second simulation was performed on the dataset used by Newell et al. (2008) that comprised of four apartments to be rented by students of New South Wales, Australia (see Dataset 2 in Appendix A for attribute information). The apartments A–D were defined by 10 attributes (like ‘Crime rate of area’). For example, [”Security of building”, 8.95, 1, 1, 1, 1] means ‘Security of building’ is an attribute with a weight of 8.95 (on a scale of 10) and Apartment A is not secure, Apartment B is secure, Apartment C is not secure and Apartment D is secure. The percentage of times each choice is made for n (1–10) attributes was calculated. Table 2 lists the associated probabilities of the attributes used in the second simulation. 3. Results 3.1. Simulation 1 The correct choice predicted by both TALLY (with nine positive attributes) and WADD (with 116.4 as the aggregate weight) is Hatsdun. When all the 12 attributes for the four cars were presented in the unconscious thought condition, the
647
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
correct choice (Hatsdun) was made 58% of the time in the Dijksterhuis et al. (2006) study, 43% of the time in Newell et al. (2008) study and 63% in Rey et al. (2009). The simulation results show that with just two attributes, Hatsdun gets selected 49% of the time and with three attributes 57% of the time (Fig. 1d) when random sampling is performed followed by TALLY (Fig. 1d). In probability-based sampling followed by both WADD and TALLY, the correct choice (Hatsdun) gets chosen around 60% of the time with just three or four attributes (Fig. 1a and b). The comparison of results from probability-based sampling and WADD with the previous empirical studies is shown in Table 3 (also see Fig. 1). 3.2. Simulation 2 With the second data set (see Dataset 2 in Appendix A), the correct choice predicted by WADD is Apartment B and correct choice predicted by TALLY is Apartment A (considering all the attributes). When all the 10 attributes for the four apartments were presented in unconscious thought condition in the Newell et al. (2008) study, Apartment B was chosen around 65% of the time and Apartment A was chosen around 17% of the time indicating that WADD is the more probable decision strategy compared to TALLY that people used to make a choice. With probability-based sampling followed by WADD, the simulation results show that with just three attributes, the results closely match with the results from the Newell et al. (2008) study following unconscious thought (see Table 4 and Fig. 2a). Interestingly, in probability-based sampling followed by TALLY, a close fit to the data reported by Newell et al. (2008) was also found with around three attributes (see Fig. 2b). It is to be noted that Apartment A is the correct choice according to TALLY only when all the attributes and in the simulations Apartment A is selected more often compared to Apartment B only when almost all (9 out of 10) attributes are considered. Random sampling followed by either WADD or TALLY does not match the results from Newell et al. (2008) and hence we do not consider them any further in our discussion.
Fig. 1. Simulations with Dataset 1. The percentage of times each car is chosen is plotted as function of n (1–12) attributes for (a) probability-based sampling of attributes followed by WADD, (b) probability-based sampling of attributes followed by TALLY, (c) random sampling of attributes followed by WADD, and (d) random sampling of attributes followed by TALLY.
Table 3 Comparison of behavioral results with data set 1 and simulation results from the probability-based sampling followed by WADD condition with three and four attributes. Simulation results indicate the percentage of times each option is chosen.
HATSDUN KAIWA DASUKA NABUSI
Experiment 1, Dijksterhuis et al. (2006)
Experiment 3, Newell et al. (2008)
Rey et al. (2009)
Simulation results (three attributes)
Simulation results (four attributes)
58 Not specified Not specified 25
43 27 27 3
63 23 3 10
56.72 23.43 10.59 0.24
63 29.65 7.27 0.07
648
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
Table 4 Comparison of behavioral results with data set 2 and simulation results from the probability-based sampling followed by WADD condition with three and four attributes. Simulation results indicate the percentage of times each option is chosen.
Apt. Apt. Apt. Apt.
A B C D
Experiment 1, Newell et al. (2008)
Experiment 2, Newell et al. (2008)
Simulation results (3 attributes)
Simulation results (4 attributes)
17.4 69.6 4.3 13.0
4.3 65.2 4.3 26.1
7.44 61.79 3.01 27.74
10.97 62.1 5.21 21.71
Fig. 2. Simulations with Dataset 2. The percentage of times each apartment is chosen is plotted as function of n (1–10) attributes for (a) probability-based sampling of attributes followed by WADD, (b) probability-based sampling of attributes followed by TALLY, (c) random sampling of attributes followed by WADD, and (d) random sampling of attributes followed by TALLY.
4. Discussion The results of our simulations support our hypothesis that sub-sampling can lead to correct choices and there is no need to posit an infinite capacity for unconscious thought (Dijksterhuis & Nordgren, 2006). This can be seen with the match between simulation results and results from empirical studies (Dijksterhuis et al., 2006; Newell et al., 2008; Rey et al., 2009). The studies on unconscious vs. conscious thought have not estimated directly the number of attributes actually used in making a choice (Acker, 2008; Dijksterhuis et al., 2006; Lassiter, Lindberg, Gonzalez-Vallejo, Bellazza, & Phillips, 2009; Newell et al., 2008; Thorsteinson & Withrow, 2009). The simulation results show that it is less likely that all or most attributes are considered during unconscious though since with all or most attributes the correct choice gets picked all or most of the time. The actual behavioral performance ranges only between 50% and 70% in most of the studies and around three or four attributes are enough to produce comparable performance. We do not argue that capacity limitations are always present for unconscious processing, in general. However, it appears that when a complex decision making problem is presented and participants are not allowed to consciously deliberate, they may still focus on a small set of attributes to make a choice. Although from simulation 1 it is not possible to estimate when the weights are used (in the sampling process or in the evaluation process, where WADD is performed), the results of simulation 2 (Fig. 2) indicate probability-based sampling (where weights of the attributes drive the initial choice of attributes) produces results closer to those from human participants (reported by Newell et al., 2008) compared to random sampling. Thus, the weights of the attributes appear to guide the initial choice or selection of attributes. The simulations also tried to explore the strategies used to combine information from different attributes. Dijksterhuis and Nordgren (2006) had argued against the strict use of WADD in unconscious thought because of the computational
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
649
complexity of WADD even though the results of unconscious thought processes match well with what WADD would predict. This argument was based on the assumption that information from all attributes was combined in unconscious thought. Our simulation results with sub-sampling make WADD a definite possibility. While WADD might be complex to perform with all the attributes, it can still be computed easily with a small subset of attributes even under the deliberation-without-attention condition. If a subset of attributes are used, then although one decision strategy predicts a correct choice (as WADD predicts Apartment B in the Newell et al., 2008 study) it could be possible that some other strategy like TALLY is used because both strategies could give the same pattern of results. This can be observed from both simulation 1 (Fig. 1a and b) and simulation 2 (Fig. 2a and b), where both WADD and TALLY followed by probability-based sampling gives almost the same results for 3–4 attributes. Thus, even though people make the choice predicted by WADD in the Newell et al. (2008) study (note that whether people used WADD or TALLY cannot be estimated in Dataset 1 because both strategies predict the same choice: Hatsdun), that might not necessarily mean that people did indeed use WADD (assuming probability-based sub-sampling). It is also possible that other simpler adaptive strategies could be used to make choices with a small set of attributes (Newell, 2005; Rey et al., 2009). Moreover, if unconscious thought (without deliberate attention) does indeed focus on a subset of attributes (like conscious thought with attention), then this potentially can explain the failures to replicate the advantages of unconscious thought over conscious thought (Acker, 2008; Newell et al., 2008; Thorsteinson & Withrow, 2009). The subset of attributes that drives a decision process might differ for different datasets. The number of attributes selected will depend on the weights of individual attributes (based on the experience of the individual). Pre-existing biases and heuristics will have a larger role to play given that a smaller number of attributes are selected. Using a small sample of relevant (or important) cues/attributes to arrive at a ‘moderately good’ judgment is consistent with the satisficing heuristic (Simon, 1990). Heuristics is commonly portrayed as an alternative to complex algorithms like the WADD. Our idea of sub-sampling of information followed by the use of algorithms like WADD/TALLY lies midway between simple heuristics like Take the Best (Gigerenzer, Todd, & the ABC Research Group, 1999) and complex algorithms which assume that decision makers considers all the information. The results suggest decision makers opt for strategies that reduce processing (Shah & Oppenheimer, 2008) due to cognitive constraints and limited attentional resources. These to some extent may explain differences obtained with different studies studying multi-attribute decision making with unconscious thought. Rey et al. (2009) describe two opposing views (and predictions made): the powerful unconscious view and the conscious attention view. According to the powerful unconscious view, only a small number of attributes are considered in the ‘conscious thought condition’ and many/all attributes are considered in the ‘unconscious thought condition’ (Dijksterhuis & Nordgren, 2006). In contrast, the conscious attention view assumes that only a few attributes are considered in the ‘unconscious thought condition’ (where the amount of time allocated to processing is less) and many/all attributes are considered in the ‘conscious thought condition’ (where processing time is more). Rey et al. (2009) found a significant difference in decision making performance only between immediate thought and conscious thought. They argued that this could be due to the fact that with increase in the number of categories (in the conscious thought condition), the difference between the cars (calculated from the mean evaluation score = Mean weight of attribute/Number of attributes sampled) reduces as the number of attributes increase. This line of reasoning was based on the assumption that the sampling process strictly selects the attributes according to a ranked order in terms of weights. However, this might not be the case due to potential individual differences in attribute selection. The simulations in the current study used probability-based sampling and not a direct analysis of the dataset as done in Rey et al. (2009) to account for such individual differences. When probability-based sampling is performed on the attributes a large number of times, the difference between the cars do not reduce as much as indicated in the analysis by Rey et al. (2009). Fig. 3 shows the aggregate mean evaluation score (=(Sum of weights/Number of trials)/Number of attributes) of all the cars as a function of number of attributes sampled (using probability-based sampling). The difference between the best car (Hatsdun) and the second best (Kaiwa) does not change drastically as more number of attributes is considered for both WADD (Fig. 3a) and TALLY (Fig. 3b). Similar to Rey et al. (2009), the difference between Hatsdun and Kaiwa does reduce with increase in the number of attributes. For example, with WADD the difference is negatively correlated, r = .921, t(1, 10) = 7.456, p = .000021. The current simulations capture the idea of bounded rationality in the sampling process and assume operation of a rational strategy like (WADD or TALLY) after the attributes are sampled. The suggestion of Rey et al. (2009) supposes that a decision maker rationally selects the attributes based on the weights (based on weights) but after the attributes are selected, the decision taken is not purely rational (where the best choice is not selected always if the evaluated values of the choices are close to each other). However, it seems reasonable to assume that attributes might not always be selected in strict rank order and there is also no (simple) algorithmic way that could explain how a decision is arrived at when the differences between the alternatives are close to one another. Both the powerful unconscious and the conscious attention views treat attention and consciousness almost synonymously but it is quite possible that attention (and hence deliberation) could be manipulated independently of consciousness. Different types of attention or differences in allocation of attentional resources might result in differences in decision making independent of manipulations of consciousness in thought. The amount of deliberation would depend on available attentional resources. Keeping this in mind, we propose a simple framework (see Fig. 4) in which attention/deliberation and consciousness constitute independent dimensions that could be varied in a given task. In this framework, the unconscious thought condition involves less attention and unconscious thinking, the conscious deliberation condition involves more
650
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
Fig. 3. Aggregate mean evaluation score of the cars as a function of number of attributes using (a) WADD and (b) TALLY.
Fig. 4. Schematic of the unified conscious attention view of decision making, where attention/deliberation and consciousness constitute independent dimensions.
attention and conscious thought, and the immediate thought condition may involve conscious thought but with less attention/deliberation. These conditions may also differ in terms of the number of attributes selected. The current simulations indicate that the quality of decisions improve with increase in the number of attributes considered in making a decision. However, it is not clear that just the number of attributes used determines the performance in empirical studies or other factors like memory (and experience) also play a role. Memory-related processes have been proposed to explain effects associated with unconscious thought (Lassiter et al., 2009; Shanks, 2006). Lassiter et al. (2009) have argued that conscious thought may employ memory-related processes and unconscious thought might employ online computation of choice when the attributes are presented. Our simulations do not directly address differences between conscious and unconscious thought, but do indicate that only a smaller number of attributes be memorized or processed in either mode of thinking. An accurate examination of memory (implicit and
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
651
explicit) of attribute information would enable us to determine the approximate number of attributes used in making a choice. Based on the simulation results, we conclude that in a multi-attribute decision problem involving unconscious thought without deliberation, people might consider only a subset of attributes and not all (or even most) of the attributes as suggested by UTT. Our simulations are consistent with the unified conscious attention view of decision making. Further studies and computational models would be needed to understand the selection of attributes and the way they are combined to make choices in complex decision making problem. Appendix A A.1. Dataset 1
[”Gas Mileage”, 18.3, 1, 1, 1, 1], [”Handling”, 16.5, 1, 1, 1, 1], [”Environment Friendly”, 15.6, 1, 1, 1, 1], [”Sound System”, 14.6, 1, 1, 1, 1], [”Service”, 14.3, 1, 1, 1, 1], [”Ease of Shifting Gears”, 12.9, 1, 1, 1, 1], [”Trunk Space”, 12.3, 1, 1, 1, 1], [”Legroom”, 11.8, 1, 1, 1, 1], [”New”, 10.2, 1, 1, 1, 1], [”Available in different colors”, 6.1, 1, 1, 1, 1], [”Has sunroof”, 5.9, 1, 1, 1, 1], [”Has cupholders”, 1.6, 1, 1, 1, 1]. A.2. Dataset 2
[”Security of building”, 8.95, 1, 1, 1, 1], [”Rent”, 8.60, 1, 1, 1, 1], [”Crime Rate of Area”, 8.36, 1, 1, 1, 1], [”Flatmate is friend”, 7.91, 1, 1, 1, 1], [”Size of apartment”, 7.56, 1, 1, 1, 1], [”Kindness of neighbors”, 5.41, 1, 1, 1, 1], [”View”, 5.18, 1, 1, 1, 1], [”Built-in wardrobe”, 4.70, 1, 1, 1, 1], [”Direction”, 4.61, 1, 1, 1, 1], [”Leisure facilities”, 4.59, 1, 1, 1, 1]
References Acker, F. (2008). New findings on unconscious versus conscious thought in decision making: Additional empirical data and meta-analysis. Judgment and Decision Making, 3, 292–303. Calvillo, D. P., & Penaloza, A. (2009). Are complex decisions better left to the unconscious? Further failed replications of the deliberation-without-attention effect. Judgment and Decision Making, 4, 509–517. Chong, S. C., & Treisman, A. (2005). Statistical processing: Computing the average size in perceptual groups. Vision Research, 45, 891–900. Dijksterhuis, A. (2004). Think different: The merits of unconscious thought in preference development and decision making. Journal of Personality and Social Psychology, 87, 686–698. Dijksterhuis, A., Bos, M. W., Nordgren, L. F., & van Baaren, R. B. (2006). On making the right choice. The deliberation-without-attention effect. Science, 311, 1005–1007. Dijksterhuis, A., Bos, M. W., van der Leij, A., & van Baaren, R. B. (2009). Predicting soccer matches after unconscious and conscious thought as a function of expertise. Psychological Science, 20, 1381–1387. Dijksterhuis, A., & Nordgren, L. F. (2006). A theory of unconscious thought. Perspectives on Psychological Science, 1, 95–109. Dijksterhuis, A., & van Olden, Z. (2006). On the benefits of thinking unconsciously: Unconscious thought can increase post-choice satisfaction. Journal of Experimental Social Psychology, 42, 627–631. Gigerenzer, G., & Todd, P. M.The ABC Research Group. (1999). Simple heuristics that make us smart. Oxford, England: Oxford University Press. Lassiter, G. D., Lindberg, M. J., Gonzalez-Vallejo, C., Bellazza, F. S., & Phillips, N. D. (2009). The deliberation-without-attention effect: Evidence for an artifactual interpretation. Psychological Science, 20, 671–675. Myczek, K., & Simons, D. J. (2008). Better than average: Alternatives to statistical summary representations for rapid judgments of average size. Perception & Psychophysics, 70, 772–788. Newell, B. R. (2005). Re-visions of rationality? Trends in Cognitive Sciences, 9, 11–15. Newell, B. R., Wong, K. Y., Cheung, J. C. H., & Rakow, T. (2008). Think, blink or sleep on it? The impact of modes of thought on complex decision making. The Quarterly Journal of Experimental Psychology, 62, 707–732.
652
N. Srinivasan, S. Mukherjee / Consciousness and Cognition 19 (2010) 644–652
R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. . Rey, A., Golstein, R. M., & Perruchet, P. (2009). Does unconscious thought improve complex decision making? Psychological Research, 73, 372–379. Shah, A. K., & Oppenheimer, D. M. (2008). Heuristics made easy: An effort-reduction framework. Psychological Bulletin, 134, 207–222. Shanks, D. R. (2006). Making choices without deliberating. Science, 313, 760. Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19. Thorsteinson, T. J., & Withrow, S. (2009). Does unconscious thought outperform conscious thought on complex decisions? A further examination. Judgment and Decision Making, 4, 235–247.
Consciousness and Cognition 19 (2010) 653–655
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Commentary
Consciousness is still in business q Yossi Guterman * Department of Psychology, Tel-Hai Academic College, D.N. Galil Elion 12210, Israel
a r t i c l e
i n f o
Article history: Received 10 February 2010 Available online 8 April 2010 Keywords: Working memory Consciousness Non-conscious processing Executive functions Pattern extraction Strategy
a b s t r a c t In a recent study (Hassin, Bargh, Engell, & McCulloch, 2009, Exp. 4) half of the participants were informed of the occasional occurrence of location regularities (patterns) in visual stimulus sets, while the other half was not. Evidence was presented to the effect that uninformed participants extracted the patterns from the displays better than the informed participants. The authors interpret their finding as demonstrating that working memory (WM) can operate non-consciously. However, inspection of the data suggests that rather than being more effective than the informed participants in extracting patterns, uninformed participants were more strongly affected by the ‘‘Broken Patterns” that served as misleading cues. Thus whereas the findings may support the possibility of non-conscious operation of low level WM functions, they nevertheless underscore the importance of conscious awareness as far as higher level functions are concerned. Ó 2010 Elsevier Inc. All rights reserved.
A recent study by Hassin and colleagues (2009) attempted to demonstrate the power of non-conscious processing in working memory (WM). The researchers engaged the participants in a task in which a fast decision, whether a disk appearing in changing locations on a display is empty or filled, had to be made. In each block, sets that consisted of five consecutive stimulus presentations could constitute either a (complete) Pattern, a Broken Pattern, a Control Set, or a random set. In a Pattern, the disks appeared in locations that constituted a recognizable outline if connected by an imaginary line. In a Broken Pattern, four disks conformed to the pattern, but the fifth disk went out of line and appeared in an unpredictable location. In a Control Set, the first three disks appeared in random locations and the last two conformed to the pattern. Each block consisted of 10 different complete patterns, 10 related Broken Patterns, 10 related control patterns and 70 random patterns. In experiment 4, half of the participants were told in advance of the existence of patterns (Explicit condition) while the others were not (Implicit condition). RTs to the last disk in each set constituted the dependent variable which was compared across instruction conditions. The authors reported an effect for the set type manipulation in the Implicit group alone, with Patterns resulting in shortest reaction times and Broken Patterns resulting in the longest ones (Pattern 514 ms; Control 547 ms; Broken Pattern 566 ms). They interpreted their findings as indicating that only the Implicit group participants (non-consciously) extracted patterns from the display, whereas the Explicit group participants failed to do so. This was considered support for the notion of non-conscious goal-oriented working memory processing. However, inspection of their descriptive data reveals an additional finding, which was not mentioned in the text: the participants in the Explicit group, in all three non-random set type conditions (Pattern 508 ms; Control 511 ms; Broken Pattern 519 ms) were as fast as the Implicit group participants in their fastest (Pattern) sets (514 ms). A possible interpretation of these data is that conscious awareness of the existence of patterns in the displays enabled the participants to find ways to benefit from the predictability associated with complete patterns while nullifying the cost associated with the Broken Patterns or the controls (unfortunately, no data on the RTs in random sets were provided). q
Commentary on Hassin, R. R., Bargh, J. A., Engell, A. D., & McCulloch, K. C. (2009). Implicit working memory. Consciousness and Cognition, 18(3), 665–678. * Fax: +972 77 750 5014. E-mail address: [email protected]
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.005
654
Y. Guterman / Consciousness and Cognition 19 (2010) 653–655
Alternatively, participants in the conscious condition might have given up altogether on pattern identification (considering the very low ratio of valid patterns), and adopted a strategy of a ‘‘wide attentional beam”.1 Regardless of the particular type of strategy adopted, some important WM, or to be more specific, Executive functions had to be put to work, namely: Planning the strategy of disk and pattern search (e.g. narrowing or widening of the attentional ‘‘beam”). Monitoring the implementation of the strategy in terms of its feasibility (ease of execution) and outcomes (speed of responding and/or success in pattern assessment). Strategy evaluation and adjustment (e.g. assessing RT cost for Broken Patterns; inhibiting the tendency to emit an anticipation-based response after four consecutive apparently patterned locations; giving up on pattern search). In other words, to obtain superior performance, as the one shown in the Explicit condition, the involvement of higher order functions was needed: metacognition had to be coordinated with perception and spatial analysis to ensure that strategies were developed, implemented and adjusted on time. To demonstrate how WM was tapped in their paradigm, Hassin lists a series of operations that were needed to respond successfully to the pattern manipulation and that were supposedly accomplished in the Implicit condition: It requires active maintenance of ordered information for relatively short periods of time; in addition, it requires context-relevant updating of information (with incoming disks) and goal-relevant computations (i.e., pattern extraction and anticipation formation). Lastly, the information is processed in the service of current goals (of being fast and accurate), and is readily available to bias cognition and behavior (thus speeding/slowing responses) (p. 668). As argued below, all of the operations listed above represent low-level functions typically accomplished by WM slave sub-systems or by associative networks, with little involvement of the central executive as characterized by Baddeley (2003), Miyake and colleagues (2000) or Smith and Jonides (1999): Maintenance of (spatially) ordered information for a relatively short period of time is a task typically attributed to the visuospatial sketch pad in WM (Baddeley & Hitch, 1974). What is claimed to be context-relevant updating of memory appears to be context-indifferent memory updating, with all disk locations indiscriminately registered. This is evident from the effects of the Pattern-Type manipulation in the Implicit condition that could only have occurred if the participants gave equal weight to all the available location data. Goal-relevant computations to extract patterns and form anticipations could have been considered high level operations had the patterns been entirely unfamiliar to the participants. However, as reported by the authors‘, many of the patterns were familiar (e.g. Zig Zag), and thus needed to be recognized rather than computed. Pattern recognition is an operation that is typically attributed to activation of long-term memory units by stimulus configurations (Rumelhart, Hinton, & McClelland, 1986). Indeed, as Hassin points out, there had to be temporal integration of the consecutive disk locations into recognizable configurations. However, it should be noted that the stimuli within each set were visible for 500–550 ms each and were separated by 150 ms intervals, making each set last about 3.5 s. At the same time, sets were clearly demarcated from each other by 1500 ms intervals (for an interval/set length ratio of 42%). Thus temporal integration could have relied on powerful bottom-up cues which only minimally called for the involvement of higher order working memory functions. Anticipation formation, based on recognized regularities has long been accepted as a basic non-conscious operation of the perceptual system in the service of object recognition and perceptual constancies (Neisser, 1976; Rock, 1983). Anticipation-based biasing of cognition and behavior in the service of current goals has been shown to occur without central executive involvement in perceptual organization (Winkler, Denham, & Nelken, 2009), language comprehension (Otten & Van Berkum, 2009) and control of skilled motor behavior (Wolpert & Flanagan, 2001), to name a few. All of these operations may indeed be executed ‘‘non-consciously” in the sense of lack of awareness of their very occurrence or the role they play. At the same time they are limited in their ability to cope with situations in which misleading cues prime responses that are sometimes inadequate (for analogies in the attention domain see Neely (1977), Posner (1980)). The Hassin et al. (2009) study illustrates nicely the higher order regulatory role of consciousness and the limitations of non-conscious processing. By showing what non-conscious processes can do, it also demonstrates what they cannot do: think. References Baddeley, A. D. (2003). Working memory: Looking back and looking forward. Nature Reviews Neuroscience, 4, 829–839. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), The psychology of learning and motivation (pp. 47–89). New York: Academic Press. Hassin, R. R., Bargh, J. A., Engell, A. D., & McCulloch, K. C. (2009). Implicit working memory. Consciousness and Cognition, 18, 665–678. 1 Note that considering the crude discrimination required between full and empty disks, the advantage of pattern identification could have been minimal compared to a ‘‘wide beam” strategy. This possibility is supported by the absence of differences in the Implicit group between the Pattern and Control conditions in the first three experiments.
Y. Guterman / Consciousness and Cognition 19 (2010) 653–655
655
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive functions and their contributions to complex ‘‘frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41, 49–100. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226–254. Neisser, U. (1976). Cognition and reality: Principles and implications of cognitive psychology. San Francisco, CA: Freeman. Otten, M., & Van Berkum, J. J. A. (2009). Does working memory capacity affect the ability to predict upcoming words in discourse? Brain Research, 1291, 92–101. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32, 3–25. Rock, I. (1983). The logic of perception. Cambridge, MA: MIT Press. Rumelhart, D. E., Hinton, G. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations in the microstructure of cognition. Cambridge, MA: MIT Press. Smith, E. E., & Jonides, J. (1999). Storage and executive processes in the frontal lobes. Science, 283, 1657. Winkler, I., Denham, S. L., & Nelken, I. (2009). Modeling the auditory scene: Predictive regularity representations and perceptual objects. Trends in Cognitive Sciences, 13, 532–540. Wolpert, D. M., & Flanagan, J. R. (2001). Motor prediction. Current Biology, 11, 729–732.
Consciousness and Cognition 19 (2010) 656–666
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Individual differences in metacontrast masking are enhanced by perceptual learning Thorsten Albrecht *, Susan Klapötke, Uwe Mattler Georg-Elias-Müller Institute for Psychology, Georg-August University, Göttingen, Germany
a r t i c l e
i n f o
Article history: Received 20 July 2009 Available online 30 December 2009 Keywords: Perceptual learning Individual differences Priming Metacontrast masking Consciousness
a b s t r a c t In vision research metacontrast masking is a widely used technique to reduce the visibility of a stimulus. Typically, studies attempt to reveal general principles that apply to a large majority of participants and tend to omit possible individual differences. The neural plasticity of the visual system, however, entails the potential capability for individual differences in the way observers perform perceptual tasks. We report a case of perceptual learning in a metacontrast masking task that leads to the enhancement of two types of adult human observers despite identical learning conditions. In a priming task both types of observers exhibited the same priming effects, which were insensitive to learning. Findings suggest that visual processing of target stimuli in the metacontrast masking task is based on neural levels with sufficient plasticity to enable the development of two types of observers, which do not contribute to processing of target stimuli in the priming task. Ó 2009 Elsevier Inc. All rights reserved.
1. Introduction One approach to understanding visual perception is to examine participants’ ability to discriminate stimuli in conditions with limited sensory input. Psychophysical studies have used masking procedure to limit the sensory input and to examine stimulation parameters that determine participants’ performance in perceptual tasks (e.g., Bachmann, 1984, 1994; Breitmeyer & Ögmen, 2006; Turvey, 1973). However, also using masking, priming studies have demonstrated processing of subliminal stimuli that participants cannot discriminate (e.g., Eimer & Schlaghecken, 1998; Fehrer & Raab, 1962; Klotz & Neumann, 1999; Marcel, 1983; Mattler, 2003; Neumann & Klotz, 1994; Schmidt, 2000, 2002; Vorberg, Mattler, Heinecke, Schmidt, & Schwarzbach, 2003). Interestingly, this literature does not address individual differences because it focuses on general principles of perception that apply to a large majority of participants. However, due to the neural plasticity in the visual system, differences in individual perceptual experience are potentially capable of modifying observers’ performance in perceptual tasks that could result in stable individual differences in performing specific tasks. Perceptual learning ranges from long lasting effects of a single exposure to a stimulus to the improved ability to perform specific perceptual tasks after substantial practice (Fahle & Poggio, 2002). In addition to differences in the amount of practice, stable individual differences might also result when specific predispositions render observers more efficient in stimulus processing at certain levels in the visual system. This view is consistent with current theories of perceptual learning that assume multiple potential levels in the visual system at which perceptual learning can take place (Yotsumoto & Watanabe, 2008). For instance, the Reverse Hierarchy Theory (RHT, Ahissar & Hochstein, 1993, 1997) assumes two phases in perceptual learning: an initial attentional phase in which the appropriate processing level is determined, and a following phase of plasticity in
* Corresponding author. Address: Georg-Elias-Müller Institute for Psychology, Georg-August University Göttingen, Gosslerstr. 14, D-37073 Göttingen, Germany. Fax: +49 551 393662. E-mail address: [email protected] (T. Albrecht). 1053-8100/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2009.12.002
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
657
which the neural structure at the selected level changes. This theory could explain individual differences in perceptual tasks by assuming that individuals can select different processing levels to perform a task. A similar prediction has been made by ecologically motivated accounts of perceptual learning, which assume individual differences occur due to participants’ use of different informational variables (e.g., Jacobs, Runeson, & Michaels, 2001; Runeson & Andersson, 2007; Withagen & van Wermeskerken, 2009). Here we report a case in which qualitative individual differences developed and stabilized spontaneously despite identical learning conditions. We used a metacontrast masking paradigm, in which the target stimulus is followed by a mask, whose contours are displayed contiguous to the contours of the target (Fig. 1a and b). Under these conditions target visibility is a function of the stimulus onset asynchrony (SOA) between the target and the mask. Two types of masking functions have been traditionally distinguished. With Type A functions target visibility is minimal at short SOAs and increases with increasing SOA. With Type B functions target visibility follows an U-shaped function with minimal target visibility at intermediate SOAs (Kolers, 1962). According to previous research, the type of the masking function is determined by the relation between the duration and/or intensity of mask and target stimuli (Breitmeyer & Ögmen, 2000), and by the spatial layout of the stimuli (Duangudom, Francis, & Herzog, 2007; Francis & Cho, 2008; Francis & Herzog, 2004). Attentional effects have been found to affect the level rather than the type of the masking function (Ramachandran & Cobb, 1995; Shelley-Tremblay & Mack, 1999). However, as an exception to these findings, Weisstein (1966) reported that the peak of Type B masking functions shifted towards smaller SOAs when attention had to be divided across several spatial positions rather than being focused on one position alone. Type B masking functions have typically been found in brightness rating tasks, whereas Type A masking functions result from simple detection and speeded response time tasks (for a review, see Breitmeyer & Ögmen, 2006). Task dependent differences in metacontrast masking have been explained by the assumption that participants apply different criterion contents when they perform different tasks. Criterion content refers to the stimulus attribute, psychological dimension or perceptual
Fig. 1. Trial sequence and stimuli. (a) Sequence of Events. (b) Target and masking stimuli.
658
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
cue a judgment is based on (Ventura, 1980). A change of criterion content has also been assumed to account for the finding that Type B metacontrast masking can be reduced or even absent after several sessions of practice (Hogben & DiLollo, 1984; Ventura, 1980). Ventura (1980), for instance, has used a stimulus sequence that elicited a biphasic perception with an initial bright phase and a following dim phase. Type A or Type B masking functions were found when brightness ratings were based on the initial or second perceptual phase, respectively. With increasing practice, Type B masking was reduced and participants reported a shift of criterion content from the second to the initial perceptual phase. Changes in criterion contents might also account for the finding that Type A or Type B masking functions have been found when participants were instructed to respond fast or slow, respectively (Lachter & Durgin, 1999; Lachter, Durgin, & Washington, 2000). According to Lachter and colleagues, fast responses are based on stimulus information that is available briefly after stimulus presentation whereas slow responses are based on stimulus information that is available only at a later stage of processing. Therefore, Type B masking functions on trials with slow responses indicate a failure to retain information over short periods of time (Lachter et al., 2000). Thus, the concept of criterion content predicts individual different masking functions if one assumes individually different predispositions for specific criterion contents. We found that individuals who performed the same perceptual task showed a predisposition to either Type A or Type B masking from the beginning on, and progressively segregated into groups with either Type A or Type B masking functions. 2. General methods 2.1. Participants Sixteen students (four male) from Göttingen University between 18 and 23 years old participated in Experiments 1 and 2. Ten of them also participated in Experiments 3 and 4. All had normal or corrected-to-normal vision and received partial course credit. 2.2. Tasks Participants were repeatedly examined in four experiments which employed stimulus sequences exemplified in Fig. 1a. In the target discrimination task participants were to respond as accurate as possible, and without speed stress, to the shape of the target stimulus (square or diamond) with a left or right hand response (Experiments 1, 3, and 4). In the choice reaction time task participants responded to the masking stimulus (square or diamond) with a left or right hand response (Experiment 2). 2.3. Stimuli The stimuli used throughout the experiments were small filled squares and diamonds (targets) subtending 1.5° of visual angle and bigger framed stimuli (masks) with square- and diamond-shaped outer contours subtending 2.6° of visual angle. The outer contours of the targets fitted neatly into the inner contours of the masks leaving a space of one pixel, which corresponds to 0.02° of visual angle (Fig. 1a and b). All stimuli were black (0.03 cd/m2) on a light gray background (72.3 cd/m2) in the center of the screen with durations of 24 ms and 108 ms for targets and masks, respectively. Targets were always presented before the mask with an SOA of 24, 36, 48, 60, 72, or 84 ms. In half of the trials the target and mask stimuli were congruent (both stimuli were squares or diamonds), in the other half of the trials, the target and mask stimuli were incongruent (one stimulus a square and the other a diamond). The congruency varied randomly across trials (Fig. 1b). Auditory feedback (1000 Hz, 100 ms) was given on each error response. 2.4. Procedure The sequence of events was the same for all experiments: Each trial started with a fixation cross for 750 ms followed by the target and then the mask (Fig. 1a). The inter-trial interval varied between 800 ms and 1850 ms following a quasiexponential distribution. For the target discrimination task participants were instructed to keep their gaze on the fixation cross throughout the trial, and to respond as accurately as possible to the shape of the target stimulus without paying attention to the masking stimulus. The participants pressed the left button upon seeing a square and the right button when seeing a diamond. Participants had to respond within 3 s after mask onset. For the choice reaction time task they were instructed to respond as fast as possible to the outer contour of the mask and to ignore the shape stimulus presented before the mask. The data of Experiments 1 and 2 was collected across two successive sessions each, and the data of Experiments 3 and 4 across a single session each. Each session was run at a separate day and comprised 13 blocks of 48 trials each. The first block of each session was considered warm-up and discarded from further analysis. Independent variables Congruency (congruent vs. incongruent) and SOA (24, 36, 48, 60, 72, and 84 ms) were varied pseudo-randomly within each block so that each of the 12 combinations was repeated 4 times in each block and 48 times in each session. In Experiment 4 only two SOAs with 24 ms and 72 ms were employed.
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
659
2.5. Data analysis Discrimination performance for each participant and SOA was assessed by a signal detection analysis resulting in measures of d’. Measures of d’ were calculated for each masking stimulus separately, and averaged across type of masking stimuli (Vorberg, Mattler, Heinecke, Schmidt, & Schwarzbach, 2004). However, to facilitate the readability of the figures we report percent correct in the figures below. Perceptual learning was analyzed based on an estimation of the performance in a certain training period by summarizing performance measures across the trials of each quarter of a session (three blocks of experimental trials, 144 trials total). In this way, perceptual learning across two sessions could be captured across eight training periods, constituting the independent variable Practice (T1–T8). The analysis of reaction time (RT) data in Experiment 2 was based on correct trials only; error data was arc-sine-transformed before analysis. RTs on correct trials were sorted for each condition and summarized by trimmed means (10% per participant and condition, Wilcox, 1997). When appropriate, Huyn–Feldt corrected p-values are reported. To facilitate readability uncorrected degrees of freedom are reported. 3. Experiment 1 3.1. Results Fig. 2a shows the performance of each participant as a function of SOA in the second session of Experiment 1. Visual inspection of the data reveals that in ten participants target visibility decreased with increasing SOA reaching a minimum at 60 ms SOA, and it increased with longer SOAs. In contrast, in five participants target visibility increased with increasing SOA across the entire range of SOAs. A single participant (Participant 14) did not show one of these patterns because in this participant target visibility did not change with SOA. Apart from this participant, all other participants either showed clear Type A masking with a minimum in visibility at a SOA of 24 ms, or clear Type B masking with a minimum in visibility at intermediate SOAs. To classify participants according to their masking functions, we conducted two types of cluster analyzes on the standardized individual masking functions based on the data of Session 2: Agglomerative hierarchical clustering with Euclidean distances as measure of distance and Ward’s linkage method (Ward, 1963), and k-means cluster algorithms. Both types of analyzes yielded exactly the same solutions. The two-group solution yielded the highest values in the Calinski–Harabasz-Criterion (Calinski & Harabasz, 1974) with VRC = 34.5 compared to a VRC = 25.4 and VRC = 23.7 for three and four group solutions, respectively. Both, agglomerative and k-means clustering revealed the same two clusters: Participant 1, 2, 3, 5, 7, 9, 11, 12, 13, and 16 – who showed Type B masking – were members of one cluster, whereas participant 4, 6, 8, 10, 14, and 15 – who showed Type A masking – were members of the second cluster. Because participant 14 was a member of the same group in both cluster analyzes, we consider this participant a member of the Type A cluster in all following analyzes, although his masking function did not look like a typical Type A masking function. Note that the same pattern of results was found when the data of this participant was excluded. Fig. 2b shows the effect of Practice across the two sessions of Experiment 1. A 3-way ANOVA with factors Group (Type A vs. Type B), SOA (24–84 ms), and Practice (training periods T1–T8) revealed that Type A observers performed better than Type B observers (main effect of Group: F(1, 14) = 48.9, p < 0.0001), and that both groups showed different masking functions (Interaction Group SOA: F(5, 70) = 41.6, p < 0.0001). Furthermore, there was a clear learning effect across the eight training periods (main effect of Practice: F(7, 98) = 18.0, p < 0.0001). Most important, however, learning was different for Type A and Type B observers (interaction Group Practice: F(7, 98) = 4.35, p = 0.001). This interaction was further modulated by SOA (interaction Group Practice SOA: F(35, 490) = 1.68, p = 0.01). To assess the differences in learning in more detail, we computed the following pairwise comparisons: For both groups of observers and each level of SOA we compared the performance at T1 with performance at each of the following training
Fig. 2. Results. (a) Individual masking functions of all participants in Experiment 1, Session 2. (b) Development of mean masking functions in Experiment 1 across four quarters of Session 1 (T1–T4) and Session 2 (T5–T8). Filled and open symbols denote Type A and Type B observers, respectively.
660
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
periods (i.e., T1 vs. T2, T1 vs. T3, and so on) using Fisher’s LSD as a rather liberal test for multiple comparisons (see e.g., Shaffer, 1995). As shown in Fig. 2b, Type A observers’ performance was better at T8 than at T1 for all SOAs (all ps < 0.01). Interestingly, performance of Type A improved initially in conditions with long SOAs and only later in conditions with short SOAs: With a SOA of 60 ms a significant improvement started already at T3 (p < 0.001), with a 72 ms SOA at T2 (p = 0.05), and with a 84 ms SOA at T3 (p = 0.05). In contrast, with a 24 ms SOA a significant performance improvement did not start before T5 (T2–T4: ps > 0.80; T5: p = 0.02). With SOAs of 36 ms and 48 ms the first significant improvement was observed at T4 and T5, respectively. On the other hand, Fig. 2b also shows that performance of Type B observers improved initially in conditions with short SOAs. With long SOAs, in contrast, we found no significant improvement during two sessions of practice: With a 24 ms SOA performance improved already from T1 to the next point in time T2 (p = 0.02), with a 36 ms SOA performance did not improve before T5 (p = 0.02), and with longer SOAs Type B observers’ performance did not change during the eight training periods we examined (all ps > 0.19). 3.2. Discussion The retrospective analysis of perceptual learning revealed that target visibility improved with increasing practice at different SOAs in two distinct groups of observers resulting either in Type A or in Type B masking functions. Moreover, evidence for initial perceptual learning was found at different SOAs in the two groups of observers: Performance of Type A observers initially improved with long SOAs whereas performance of Type B observers improved with short SOAs. Overall, Type A observers performed better than Type B observers. This result is at odds with models of backward masking, which predict that Type B masking is generally weaker than Type A masking (Francis, 2003; Francis & Herzog, 2004). However, because the present finding is based on group differences, it is unclear whether this effect results from different performance levels in the two groups of participants or from the underlying mechanisms of backward masking. As suggested by an anonymous reviewer, there is a possibility that the different masking functions result because the inter-trial-interval (ITI) was too short for at least some participants to prepare the processing of the target or to proceed with a specific processing strategy. Note that the ITI varied randomly between trials with 800 and 1850 ms before the warning stimulus was presented for 750 ms. Thus, there was an interval of 1550–2600 ms from the last response to the next target presentation that could be used for preparatory processes. However, no participant indicated that the time between trials was too short when we asked them in our debriefing procedure after each experimental session. Moreover, additional analyzes of our data did not provide evidence for an effect of ITI: When we compared the performance of the two groups of observers in those 25 percent of the trials with shortest ITIs (mean 847 ms) and those 25 percent of the trials with the longest ITIs (mean 1518 ms) we did not find a main effect of ITI (F(1, 14) = 1.42, p = 0.25) nor any interaction of ITI and any other independent variable (all Fs < 1, ps > 0.45). Thus, although we think it is unlikely that the short ITIs contributed to the present findings, it seems reasonable to address this issue in future research. According to Lachter and Durgin, Type A and Type B masking functions result from fast and slow responses, respectively (Lachter & Durgin, 1999; Lachter et al., 2000). However, an analysis of the RTs of the two groups of observers yielded no difference between mean RTs (973 ms and 913 ms for Type A observers and Type B observers, respectively; t(14) = 0.64, p = 0.54). To the extent that our paradigm can be compared to Lachter and Durgin’s paradigm, following Lachter and Durgin’s perspective, one would have to assume that Type A observers retained early stimulus information but Type B observers did not, although there is no difference in response speed between groups. Otherwise, our data provide no evidence for the view that the two types of observers differ in respect of retention failure, because mean RTs of both groups are comparable to those of Lachter and Durgin’s participants in slow response conditions (Lachter & Durgin, 1999). Our practice effects differ from those of previous studies (Hogben & DiLollo, 1984; Ventura, 1980), which reported a reduction of Type B metacontrast masking when practice increased. For instance, in a brightness judgment task, Ventura (1980) initially found clear Type B masking functions that were successively reduced during the course of up to five short practice sessions. This effect has been attributed to a change of criterion content, because participants reported that they changed cues provided by the stimulus on which they based their brightness judgment. In contrast, however, in our participants practice enhanced individual differences in masking functions. This finding suggests that our participants from the beginning on had certain predispositions to either one of two cues, and they had difficulties or were unable to switch between cues. Instead, perceptual learning led to more efficient processing with the preferred cues. Further research is needed to understand the experimental variables which determine whether practice effects abolish or stabilize metacontrast masking functions and whether participants can change criterion content or not. 4. Experiment 2 4.1. Method To investigate whether the perceptual learning effects found in Experiment 1 generalize across different behavioral tasks, we conducted Experiment 2 to examine the difference in the two groups based on Type A or Type B masking functions in Experiment 1, when an independent task was performed on the same visual stimuli as in Experiment 1. Stimulus conditions remained the same (Fig. 1a and b) but the participants’ task changed: In a speeded choice reaction time task participants
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
661
responded to the shape of the masking stimulus (square or diamond) with a left or right hand response. Each participant was run in two successive sessions. Error feedback was given on each incorrect trial. 4.2. Results and discussion Overall, mean RT of the two groups of observers did not differ significantly (main effect of Group: F(1, 14) < 1, p > 0.8). An analysis of the data of each quarter of the two sessions revealed that there was a significant practice effect on mean RTs (F(7, 98) = 13.18, p < 0.0001) with faster responses in later training periods (475 ms, 460 ms, 458 ms, 455 ms, 429 ms, 432 ms, 429 ms, and 426 ms for T1–T8, respectively). This practice effect did not differ in the two groups of observers as indicated by the non-significant interaction of Group Practice (F(7, 98) < 1, p > 0.8). To determine whether the target stimulus affected the response to the mask, the priming effect was calculated as the mean RT on incongruent trials minus mean RT on congruent trials. Fig. 3 shows that both groups exhibited typical priming effects that increased monotonically with SOA in both groups (e.g., Mattler, 2003; Vorberg et al., 2003). This effect was confirmed by the significant interaction of SOA and Congruency (F(5, 70) = 43.8, p < 0.0001). The mean slope of the priming function was s = 1.21, which did not differ significantly from unity (t(15) = 1.54, p = 0.15). The priming effect differed between groups as shown by the significant interaction of Group and Congruency (F(1, 14) = 4.22, p = 0.05), with a slightly larger priming effect in Type A observers (56 ms vs. 38 ms). The priming effect was not modulated by Practice as reflected in the non-significant interactions of Practice with Congruency or any other independent variable (all Fs < 1, ps > 0.70). Most important, however, although SOA differentially modulated the perception of the effective stimulus in the two groups (Experiment 1), the priming effect was modulated by SOA in both groups in about the same way: The 3-way interaction of Congruency, SOA and Group was not significant (F(5, 70) = 1.49, p = 0.23). Analysis of choice error rate (mean 2.8%, Table 1) revealed a significant main effect for SOA (F(5, 70) = 17.53, p < 0.001) and a significant interaction SOA Congruency (F(5, 70) = 3.86, p < 0.01) reflecting an increased priming effect with increasing SOA. All other effects did not reach significance (all F < 1, p > 0.5). These findings are further evidence for the dissociation between priming effects and the visibility of the effective stimuli (Fehrer & Raab, 1962; Klotz & Neumann, 1999; Mattler, 2003, 2005, 2006; Mattler & Fendrich, 2007; Neumann & Klotz, 1994; Schmidt, 2000, 2002; Vorberg et al., 2003). Most importantly, however, this finding shows that perceptual learning in Experiment 1 was at least somewhat task specific. Therefore, perceptual learning in Experiment 1 did not result in generally modified priming-related processing of the target stimuli in each of the two groups. Instead, perceptual learning improved the processing of specific cues generated by the target-mask sequence that were differentially used by individuals of the two groups to perform the target discrimination task (Ahissar & Hochstein, 1993). The slope of the priming effects in the speeded choice RT task approached unity, which means that the priming effect increased by about 10 ms when the SOA increased by 10 ms. This replicates previous findings in favour for the view, that the target stimulus activated the corresponding motor response leading to facilitated responses on congruent trials and delayed responses on incongruent trials (Mattler, 2003; Schmidt, 2002; Vorberg et al., 2003). Target induced response activation is also suggested by electrophysiological findings which show target related response activation in the motor cortex
Fig. 3. Mean RT in the choice RT task of Experiment 2 for Type A (filled symbols) and Type B observers (open symbols) on congruent (dashed line) and incongruent trials (solid line).
662
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
Table 1 Mean error percentages in Experiment 2. Prime–mask congruency
Prime–mask SOA (ms) 24
36
48
60
72
84
Type A observers Congruent Incongruent
1.6 1.9
1.4 1.9
0.9 2.1
0.7 4.2
1.0 5.0
0.7 13.2
Type B observers Congruent Incongruent
1.3 2.2
1.4 1.4
1.6 2.2
1.7 3.8
1.8 5.6
1.5 8.1
(e.g., Leuthold & Kopp, 1998). Moreover, as demonstrated by the priming effects on accuracy data in Experiment 2, target induced response activation can lead to the execution of incorrect responses (e.g., Mattler, 2003; Vorberg et al., 2003). For instance, an experiment of Vorberg and colleagues (2003) revealed increasing error rates with increasing SOA and up to 60% erroneous responses on trials with incongruent masks. If target stimuli activate the corresponding motor response in speeded choice RT tasks, one could hypothesize that the monotonic increase of prime recognition performance with increasing SOA (Type A masking functions) might also result from target induced response activation if participants are instructed to respond fast. Therefore, target induced response activation might explain previous findings which suggest a relation between masking functions and response speed (Lachter & Durgin, 1999; Lachter et al., 2000). Note, that individual differences in Experiment 1 cannot be accounted for by response activation effects, because both groups of participants responded without speed stress and produced similarly long RTs despite different masking functions. 5. Experiment 3 5.1. Method Experiment 3 was conducted to test the stability of the perceptual learning effects of Experiment 1. We recruited five participants with Type A masking functions (one participant was no longer available) and five randomly selected participants with Type B masking functions for a repetition of the target discrimination task of Experiment 1. On average, the sessions of Experiment 3 were run 105 days after the second session of Experiment 1 (range: 33–160 days). 5.2. Results and discussion Fig. 4a and b display the data of the subsample of 10 participants who continued with Experiment 3. To facilitate comparison, Fig. 4a reproduces the data of these participants from the second session of Experiment 1, and Fig. 4b shows their performance as a function of SOA in Experiment 3. Experiment 3 replicated the findings of Experiment 1 as revealed by a 2 6 ANOVA with Group and SOA as independent variables: Type A observers performed better than Type B observers (main effect Group: F(1, 8) = 152.52, p < 0.0001), and these differences were modulated by SOA, as revealed by the significant interaction of Group SOA (F(5, 40) = 31.0, p < 0.0001). The main effect for SOA did not reach significance (F(5, 40) = 1.15, p > 0.30). To compare the performance in Experiments 1 and 3 we conducted a second ANOVA with Time (Experiment 1 vs. Experiment 3) and SOA as within-subjects-factors, and Group as between subjects-factor. Like in the analysis of the data
Fig. 4. (a) Individual masking functions in Experiment 1, Session 2 for the remaining 10 participants who took part in Experiment 3. (b) Individual masking functions in Experiment 3, 33–160 days later. (c) Individual masking functions in Experiment 4 with fixed SOAs for six blocks of trials following 1–10 days after Experiment 3. Color codes different participants. Filled and open symbols denote Type A and Type B observers, respectively.
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
663
of Experiment 3 alone, the main effect of Group (F(1, 8) = 140.9, p < 0.0001) and the interaction of Group SOA was significant (F(5, 40) = 38.9, p < 0.0001), whereas no main effect for SOA was obtained (F(5, 40) = 0.87, p = 0.48). Most importantly, however, factor Time did not yield any significant main effect or interaction (always F < 2.0, p > 0.16), indicating a certain stability of participants’ performance during the period between Experiment 1 and Experiment 3 (compare Fig. 4a and b). Thus, the present learning effects seem to remain largely unchanged between experiments thus indicating that learning this perceptual task results in masking functions of a certain Type and level of expression that persist across a relatively long period of time. 6. Experiment 4 A final test of the stability of individual differences was established in Experiment 4, which examined whether participants’ perceptual performance improves if they are shown only trials with that SOA at which they had performed poor in previous sessions. We presented an entire half of the session with a constant short SOA of 24 ms and the other half of the session with a constant long SOA of 72 ms. These SOAs were chosen because they discriminated best between the two groups of observers. The short SOA was relatively easy for Type B observers but relatively difficult for Type A observers, the long SOA was relatively easy for Type A observers but relatively difficult for Type B observers. Six blocks with difficult trials were presented in the first half of the experiment followed by six blocks with easy trials in each group. Again, feedback was given on every error trial. For the different participants this session followed 1–10 days after their last session of Experiment 3. 6.1. Results The findings replicated those of Experiments 1 and 3 (see Fig. 4c): The two groups of observers performed differently at the two levels of SOA (interaction Group SOA: F(1, 8) = 89.6, p < 0.0001) with Type A observers showing a better overall performance than Type B observers (main effect Group: F(1, 8) = 139.3, p < 0.0001). Furthermore, we found a significant main effect of SOA (F(1, 8) = 8.11, p = 0.02). To examine whether participants’ performance improved due to the blocked presentation of one of two SOAs, we computed orthogonal linear contrasts between Experiment 3 and Experiment 4 for each group at each SOA. Type A observers did not improve significantly with the short SOA (F(1, 8) = 2.44, p = 0.16) but did improve with the long SOA (F(1, 8) = 13.39, p = 0.006). Type B observers improved only with the short SOA (F(1, 8) = 6.13, p = 0.03) but not with the long SOA (F(1, 8) = 0.03, p = 0.87). 6.2. Discussion Experiment 4 provides further evidence for the stability of the effects of perceptual leaning in previous sessions. Additional evidence for further improvements was found primarily in those SOA conditions that were easy for the respective group of observers but not in their difficult SOA conditions. Thus, participants further improved in conditions in which they performed well before, but did not improve in previously poor conditions. Again, no evidence for a reduction in metacontrast masking due to practice could be found. In contrast to findings of Ventura (1980) and Hogben and DiLollo (1984), which suggest that participants shifted the criterion content, our results suggest that participants were biased to use particular cues and they had difficulties using cues that they had not been using before, although other participants were able to use these other cues. 7. General discussion Participants who practiced a target discrimination task in a metacontrast masking paradigm, spontaneously and progressively segregated into groups with either Type A or Type B masking functions. A retrospective analysis of the data from two sessions revealed that individuals who showed Type A and Type B masking functions at the end of the sessions differed already in the first three blocks of the experiment, which suggests that they had predispositions to either one of the two types of perceptual responding acquired in their previous lives. Practice enhanced group differences mainly because performance in the two groups improved in different SOA conditions. These effects of perceptual learning were at least somewhat task specific because the effect of SOA did not vary across groups when participants responded to the masking stimulus in a speeded choice RT task. Two additional experiments revealed that the perceptual learning effects of the initial experiment remained relatively stable: Within groups, the masking functions did not differ in a follow-up test about 3 months later. Finally, even when participants performed the target discrimination task during 288 trials with the stimulus conditions that were most difficult for them, performance did not improve despite single trial feedback. On the other hand, when they performed the task with only those stimulus conditions that were easy for them, a further performance increase was obtained. These results show that participants have considerable difficulties switching from their preferred type of processing to become a different type of observer. To the best of our knowledge, this is the first demonstration showing that two types of metacontrast masking functions can be based on individual differences in learning despite identical learning conditions.
664
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
Individual differences have been a vexing problem in perceptual learning (Fahle & Poggio, 2002) and they have been deemed undesirable complications in backward masking (Marcel, 1983; however, see Mattler, 2003, who took advantage of individual differences). Weisstein, Jurkens, & Onderisin (1970) and Weisstein (1972) also reported considerable variation in the SOA at which masking was strongest and emphasized the importance of individual data for fitting masking functions to models. The differences in the two groups of observers reported in the present study are unlikely to be due to motivational differences, because both groups improved with practice in some stimulus conditions. However, a few studies related individual differences in perceptual tasks to measures of intelligence (for a review see Nettelbeck, 2001). Therefore, we cannot exclude the possibility that differences in general cognitive styles, like e.g., field dependence, levelers vs. sharpener, contribute to the differences reported here. This issue has to be dealt with in future research. Within the context of the Reverse Hierarchy Theory (RHT), we suspect that there were initial differences between individuals either directly in the neural response involving neuroanatomically early levels of stimulus processing, or in later processes related to the attentional or response-related selection of an adequate neural level of processing. Such a gradual initial difference between individuals could lead to the spontaneous development of a dichotomous difference between observers through some kind of ‘‘winner takes all” algorithm that exaggerates the initial bias. The categorical difference between observers, as reflected in the two distinct masking functions could then result as suggested by RHT. Ahissar and Hochstein (1993, 1997) did not explicitly discuss masking functions. However, according to the principles of their theory as we use them here two types of masking functions can result if there is a choice between two types of cues that are processed at neural levels that have sufficient plasticity to enable perceptual learning. Thus, we assume that due to an initial bias or a predisposition acquired earlier in life participants selected one of two neural levels and due to perceptual learning at the selected levels of processing they developed either to strong Type A or strong Type B observers. This view is consistent with our finding that the difference between masking functions seem to occur within the initial three blocks of trials, indicating that participants selected the neural levels of processing in the perceptual hierarchy very early before performance was further modulated by perceptual learning. To examine the initial plasticity of participants’ visual system, it would be interesting to ask experienced observers how they make their judgments and then see whether naïve participants can be trained to learn a given cue and produce Type A or Type B masking functions, or whether Type A observers change to Type B observers (and vice versa) when they use the corresponding criteria. If the two types of processing which lead to two types of masking functions emerge spontaneously, participants should have difficulties to learn both cues with the same ease. Metacontrast masking has been used previously in priming studies to demonstrate the processing of unconscious stimuli (e.g., Klotz & Neumann, 1999; Mattler, 2003; Neumann & Klotz, 1994; Vorberg et al., 2003). The present study provides new evidence for the processing of unconscious visual stimuli by showing virtually the same priming effects in participants who have learned to discriminate the targets (primes) with either Type A or Type B masking functions. These findings show that the practice effect related to the processing of the target (prime) stimuli did not modulate the effects of these stimuli in the speeded choice RT task. Therefore, it could be assumed that the functioning of neural levels of target (prime) processing that was changed in the course of the target discrimination task did not contribute to the priming effect of these stimuli in the choice RT task. This view is consistent with the distinction of different visual pathways (Milner & Goodale, 1995) as well as with the idea that conscious perception crucially depends on recurrent processing (e.g., Lamme & Roelfsema, 2000). According to both views, there are common visual processes and specific visual processes for conscious perception and motor priming. From this perspective, it seems likely that perceptual learning modulates processing of the target stimuli at a neural level that is specific for conscious perception but not necessary for motor priming. In light of a recent study, which suggests that processing in V1 is crucial for both conscious perception and unconscious priming (Sack, van der Mark, Schuhmann, Schwarzbach, & Goebel, 2009), we assume it most likely that perceptual learning in our case occurred at neuroanatomically levels later than V1. The priming effect did not change across two sessions of practice. On the one hand, this absence of practice effects might suggest that the neural levels of target (prime) processing operate effectively already after little training in the choice RT task. On the other hand, the absence of practice effects might also suggest that the priming effect results from levels of target (prime) processing that are insensitive to practice-related changes. However, no practice effects on the priming effect are consistent with RHT, which assumes that perceptual learning requires an initial top-down selection of adequate levels of processing. When the relevant information is provided by such an easily visible stimulus as the mask in the present choice RT task, RHT predicts that perceptual learning occurs only at the neural level of mask processing. Therefore, the processing of the mask should change with practice but the priming effect caused by the target (prime) should not change with practice. Our data is consistent with both of these predictions. Our findings have potentially important implications for the understanding of conscious perception. We found clear individual differences in the target discrimination task that clustered into two groups of observers. For instance, high levels of accuracy above 90% correct were obtained in Type A observers at long SOAs and low levels of accuracy below 60% correct in Type B observers. These findings suggest that the performance difference in the two groups of observers corresponds to a difference in the subjective experience of the stimulus sequence in the two groups. The data show that individuals differ in how they use cues provided by the stimulus sequence and that this difference could be a basis for differences in the conscious experiences of the stimuli. Moreover, since performance was modulated by perceptual learning, we might conclude that subjective experience of visual stimuli is modulated by perceptual learning. This interpretation is consistent with current theories of conscious perception like the global-workspace theory (Baars, 1988; Dehaene, Kerszberg, & Changeux, 1998). According to this view, one
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
665
necessary – although not sufficient – condition for the emergence of conscious experience is a certain neural activity with a certain duration which enables information to become available to a broad range of neural processes probably including those in prefrontal cortex (for a review see Maia & Cleeremans, 2005). On this background, two groups of observers with exaggerated individual differences result after sufficient perceptual learning because there are two types of neural signals which can be enhanced by participants attendance and perceptual learning leading to an increased stability in time or strength of the stimulus representation which is necessary to become available for other interpretative processes like recurrent interactions which might enable availability of the stimulus representation in the global workspace (Cleeremans, 2007). Unfortunately, in the present study we have not gathered subjective reports to test this hypothesis. Preliminary data of a recent study, however, indeed suggest different subjective experiences in Type A and Type B observers. Therefore, we think the present phenomenon might serve as a tool to study the relation between perceptual learning and conscious perception. In conclusion, we think these findings provide new evidence that might help to distinguish between models of metacontrast masking (Breitmeyer & Ögmen, 2000; Ishikawa et al., 2006) and stress the possibility that there may be no single universal model of the processes underlying masking. Moreover, the results contribute to the understanding of the priming effects of unconscious visual stimuli and to the role of conscious perception in cognition and action (Mattler, 2003). Beyond this, the phenomenon might be useful as a model for the study of neural mechanisms that are involved in the development of individual differences. Further research is need to examine the initial bias that directs perceptual learning either to Type A or Type B performance. Whether this initial bias is due to more general cognitive or attentional styles is subject of research currently under way. Acknowledgements This work was partially supported by the German Research Foundation DFG (MA 2276/3-1). We thank Robert Fendrich, Haluk Ögmen, and two anonymous reviewers for helpful comments on a previous version of this paper. References Ahissar, M., & Hochstein, S. (1993). Attentional control of early perceptual learning. Proceedings of the National Academy of Sciences, 90, 5718–5722. Ahissar, M., & Hochstein, S. (1997). Task difficulty and the specificity of perceptual learning. Nature, 387, 401–406. Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge: Cambridge University Press. Bachmann, T. (1984). The process of perceptual retouch: Nonspecific afferent activation dynamics in explaining visual masking. Perception & Psychophysics, 35, 69–84. Bachmann, T. (1994). Psychophysiology of visual masking: The fine structure of conscious experience. Commack, NY: Nova Science. Breitmeyer, B. G., & Ögmen, H. (2000). Recent models and findings in visual backward masking: A comparison, review, and update. Perception & Psychophysics, 62, 1572–1595. Breitmeyer, B. G., & Ögmen, H. (2006). Visual masking: Time slices through conscious and unconscious vision. New York: Oxford University Press. Calinski, T., & Harabasz, J. (1974). A dendrite method for cluster analysis. Communications in Statistics – Theory and Methods, 3, 1–27. Cleeremans, A. (2007). Consciousness: The radical plasticity thesis. In R. Banerjee & B. Chakrabarti (Eds.), Models of brain and mind: Physical, computational and psychological approaches (pp. 19–33). Elsevier. Dehaene, S., Kerszberg, M., & Changeux, J.-P. (1998). A neuronal model of a global workspace in effortful cognitive tasks. Proceedings of the National Academy of Sciences of the USA, 95(24), 14529–14534. Duangudom, V., Francis, G., & Herzog, M. H. (2007). What is the strength of a mask in visual metacontrast masking? Journal of Vision, 7, 1–10. Eimer, M., & Schlaghecken, F. (1998). Effects of masked stimuli on motor activation: Behavioral and electrophysiological evidence. Journal of Experimental Psychology: Human Perception and Performance, 24, 1737–1747. Fahle, M., & Poggio, T. (2002). Perceptual learning. Cambridge: MIT Press. Fehrer, M. A. E., & Raab, D. (1962). Reaction time to stimuli masked by metacontrast. Journal of Experimental Psychology, 63, 143–147. Francis, G. (2003). Developing a new quantitative account of backward masking. Cognitive Psychology, 46, 198–226. Francis, G., & Cho, Y. S. (2008). Effects of temporal integration on the shape of visual backward masking functions. Journal of Experimental Psychology: Human Perception and Performance, 34, 1116–1128. Francis, G., & Herzog, M. H. (2004). Testing quantitative models of backward masking. Psychonomic Bulletin & Review, 11, 104–112. Hogben, J. H., & DiLollo, V. (1984). Practice reduces suppression in metacontrast and in apparent motion. Perception & Psychophysics, 35, 441–445. Ishikawa, A., Shimegi, S., & Sato, H. (2006). Metacontrast masking suggests interaction between visual pathways with different spatial and temporal properties. Vision Research, 13, 2130–2138. Jacobs, D. M., Runeson, S., & Michaels, C. F. (2001). Learning to visually perceive the relative mass of colliding balls in globally and locally constrained task ecologies. Journal of Experimental Psychology: Human Perception & Performance, 27, 1019–1038. Klotz, W., & Neumann, O. (1999). Motor activation without conscious discrimination in metacontrast masking. Journal of Experimental Psychology: Human Perception and Performance, 25, 976–992. Kolers, P. A. (1962). Intensity and contour effects in visual masking. Vision Research, 2, 277–294. Lachter, J., & Durgin, F. H. (1999). Metacontrast masking functions: A question of speed? Journal of Experimental Psychology: Human Perception and Performance, 25, 936–947. Lachter, J., Durgin, F. H., & Washington, T. (2000). Disappearing percepts: Evidence for retention failure in metacontrast masking. Visual Cognition, 7, 269–279. Lamme, V. A. F., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences, 23, 571–579. Leuthold, H., & Kopp, B. (1998). Mechanisms of priming by masked stimuli: Inferences from event-related brain potentials. Psychological Science, 9, 263–269. Maia, T. V., & Cleeremans, A. (2005). Consciousness: Converging insights from connectionist modeling and neuroscience. Trends in Cognitive Sciences, 9, 397–404. Marcel, A. J. (1983). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cognitive Psychology, 15, 197–237. Mattler, U. (2003). Priming of mental operations by masked stimuli. Perception & Psychophysics, 65, 167–187. Mattler, U. (2005). Inhibition and decay of motor and nonmotor priming. Perception & Psychophysics, 67, 285–300. Mattler, U. (2006). On the locus of priming and inverse priming effects. Perception & Psychophysics, 68, 975–991. Mattler, U., & Fendrich, R. (2007). Priming by motion too rapid to be consciously seen. Perception & Psychophysics, 69, 1389–1398.
666
T. Albrecht et al. / Consciousness and Cognition 19 (2010) 656–666
Milner, A. D., & Goodale, M. A. (1995). The visual brain in action. Oxford: Oxford University Press. Nettelbeck, T. (2001). Correlation between inspection time and psychometric abilities: A personal interpretation. Intelligence, 29, 459–474. Neumann, O., & Klotz, W. (1994). Motor responses to nonreportable masked stimuli: Where is the limit of direct parameter specification? In C. Umiltà & M. Moscovitch (Eds.), Attention and Performance XV. Conscious and nonconscious information processing (pp. 124–150). Cambridge, MA: MIT Press. Ramachandran, V. S., & Cobb, S. (1995). Visual attention modulates metacontrast masking. Nature, 373, 66–68. Runeson, S., & Andersson, I. E. K. (2007). Achievement of specificational information usage with true and false feedback in learning a visual relative-mass discrimination task. Journal of Experimental Psychology: Human Perception & Performance, 33, 163–182. Sack, A. T., van der Mark, S., Schuhmann, T., Schwarzbach, J., & Goebel, R. (2009). Symbolic action priming relies on intact neural transmission along the retino-geniculo-striate pathway. Neuroimage, 44, 284–293. Schmidt, T. (2000). Visual perception without awareness: Priming responses by color. In T. Metzinger (Ed.), Neural correlates of consciousness: Empirical and conceptual questions (pp. 157–169). Cambridge, MA: MIT Press. Schmidt, T. (2002). The finger in flight: Real-time motor control by visually masked color stimuli. Psychological Science, 13, 112–118. Shaffer, J. P. (1995). Multiple hypothesis testing. Annual Review of Psychology, 46, 561–584. Shelley-Tremblay, J., & Mack, A. (1999). Metacontrast masking and attention. Psychological Science, 10, 508–515. Turvey, M. T. (1973). On peripheral and central processes in vision: Inferences from an information-processing analysis of masking with patterned stimuli. Psychological Review, 50, 1–52. Ventura, J. (1980). Foveal metacontrast: I. Criterion content and practice effects. Journal of Experimental Psychology: Human Perception and Performance, 6, 473–485. Vorberg, D., Mattler, U., Heinecke, A., Schmidt, T., & Schwarzbach, J. (2003). Different time courses for visual perception and action priming. Proceedings of the National Academy of Sciences, 100, 6275–6280. Vorberg, D., Mattler, U., Heinecke, A., Schmidt, T., & Schwarzbach, J. (2004). Invariant time course of priming with and without awareness. In C. Kaernbach, E. Schröger, & H. Müller (Eds.), Psychophysics beyond sensation. Laws and invariants of human cognition (pp. 271–288). Mahwah, NJ: Lawrence Erlbaum Associates, Publishers. Ward, J. H. Jr., (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58, 244–263. Weisstein, N. (1966). Backward masking and models of perceptual processing. Journal of Experimental Psychology, 72, 232–240. Weisstein, N. (1972). Metacontrast. In D. Jameson & L. Hurvich (Eds.). Handbook of sensory physiology (Vol. 7). Berlin: Springer. Weisstein, N., Jurkens, T., & Onderisin, T. (1970). Effect of forced-choice vs. magnitude-estimation measures on the waveform of metacontrast functions. Journal of the Optical Society of America, 60, 978–980. Wilcox, R. R. (1997). Introduction to robust estimation and hypothesis testing. San Diego, CA: Academic Press. Withagen, R., & van Wermeskerken, M. (2009). Individual differences in learning to perceive length by dynamic touch: Evidence for variation in perceptual learning capacities. Attention, Perception & Psychophysics, 71, 64–75. Yotsumoto, Y., & Watanabe, T. (2008). Defining a link between perceptual learning and attention. PLoS Biology, 6, 1623–1626.
Consciousness and Cognition 19 (2010) 667–671
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Commentary
Individual differences in metacontrast: An impetus for clearly specified new research objectives in studying masking and perceptual awareness? q Talis Bachmann Institute of Public Law and Institute of Psychology, University of Tartu, Tallinn and Tartu, Estonia
a r t i c l e
i n f o
Article history: Received 25 January 2010 Available online 23 February 2010 Keywords: Visual masking Metacontrast Perceptual learning Methods
a b s t r a c t While the majority of perceptual phenomena based research on consciousness is implicitly nomothetic, some idiographic perspective can be sometimes highly valuable for it. It may turn out that after having had a closer look at individual differences in the expression of psychometric functions a need to revise some nomothetic laws considered as the general ones arises as well. A study of individual differences in metacontrast masking published in this issue superbly illustrates this. A myriad of urgent research objectives emerges out of this study, most of them important both for clearing up the still messy theoretical picture on visual masking and for the beginning of asking whether perceptual awareness mechanisms are so universal at all. In this commentary the problems pertinent to this issue are discussed. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction Beginning with Aristotle and scrutinized by Windelband (1894), two alternative research strategies have been common to our academic knowledge culture. First, it is the nomothetic tradition aimed at finding out universal general regularities and lawful expression of traits common to all or most of the individuals studied. Second, it is the idiographic tradition that in theory says that universal rules are a simplification and in practice tries to carefully study individual and perhaps even unique traits, features and regularities of processes that characterize individuals. Consciousness science tended to be overly nomothetic – it is sufficient to thumb through the best known handbooks of consciousness or many volumes of Consciousness and Cognition to agree with this. Rarely though, some papers appear that make one think whether the mainstream nomothetic approaches to investigating consciousness processes, phenomena and mechanisms may have overlooked something important. The article by Albrecht, Klapötke, and Mattler (2010) belongs to this genre. The authors approach the traditionally nomothetic masking-research with a small ‘‘idiographic” twist and suddenly we see that the generally more or less universal picture of metacontrast functions typically dependent on variations in stimuli parameters and subjects’ tasks breaks into two pictures where in the conditions of same stimulation parameters and the same task different subjects express qualitatively different functions of masking. About one third of the subjects showed type-A, monotonic metacontrast functions and about two thirds of the subjects produced type-B, non-monotonic functions. Moreover, the bimodal qualitative picture did not weaken, but was even more emphasized with learning. The results presented by Albrecht et al. (2010) show several things we should consider seriously. If we do so, it may end up with the need to revise much of what we presently know about masking and maybe what we presently think we know about visual awareness mechanisms. q Commentary on Albrecht, T., Klapötke, S., & Mattler, U. (2010). Individual differences in metacontrast masking are enhanced by perceptual learning. Consciousness and Cognition, 19, 656–666. E-mail address: [email protected]
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.01.012
668
T. Bachmann / Consciousness and Cognition 19 (2010) 667–671
The important feature of the Albrecht et al. (2010) experimental results is that with shortest stimulus onset asynchronies (SOAs) the level of performance in type-A and type-B subjects was virtually the same. This means that change in performance-levels with increasing SOAs could not depend on the initial different level of perceptual availability and accessibility of target information, i.e., in the conditions where the mask-free time for target processing was minimal performance was comparable in the two groups. This may not be a mere coincidence, but an indirect support for the possibility that with shortest SOAs the sensory contents of the perceptual images of targets were of the same type and of the comparable quality for the both groups. This emphasizes the significance of the work by Albrecht et al. (2010). I will go through some of the possible research issues that should receive an impetus from the article by Albrecht et al. (2010). In doing so I will occasionally suggest new experiments to be carried out in order to develop this direction of research. 2. Is it a difference in the direct phenomenal experience or a difference in the criteria? The two types of masking functions can be either a result of different phenomenology or a result of different strategies of using one or another aspect or feature included in the phenomenally present target-mask perceptual image. In their main experiments Albrecht et al. (2010) asked subjects to respond without speed stress and this is where phenomenology close to invariance or high similarity between subjects would enable different strategies – responding either to the perceptual cues at an early stage of microgenesis (Bachmann, 2000) or at a later stage. Although the qualities of the percepts as they undergo change with time passing (e.g., between 50 ms post-stimulus and 500 ms post-stimulus) may be highly compatible between subjects, the strategy of whether to ground the response on early qualities or late qualities can nevertheless make a difference. In the first putative experiment I suggest we should ask is there any relation between types of metacontrast functions and attempts to focus on early or late microgenetic qualities of target-mask displays? If we consider the sizes of stimuli used by Albrecht et al. (2010) it is easy to see that compared to the sizes used in earlier numerous metacontrast studies (reviews: Bachmann, 1994; Breitmeyer & Ög˘men, 2006), the targets (1.5°) and masks (2.6°) tend to be quite large. This allows subjects to change freely the spatial focus of where in space (and related to what features) they want to analyze the phenomenal stimulus-images mentally. Follow-up experiments should test on what features, in what order, in what combination, at what level of the spatial hierarchy of features the mental scrutiny is carried out. In our lab we designed and programmed an exploratory experiment precisely replicating all spatiotemporal and energyrelated characteristics and stimuli values used by Albrecht et al. (2010) and introspected on our experiences brought about by these displays. First we noticed that contrary to some typical target-mask metacontrast displays where targets of about 0.5–0.9° and correspondingly larger masks were used, with Albrecht et al. (2010) displays it was difficult to adopt a strategy of trying to attend to the gross shape of the filled-in dark entity of the target. This strategy simply did not work so well as the other one. The other one consisted in trying to attend to the very fine sharp-white contour formed between the inner edges of the mask and outer edges of the target. This seemed to work better – square or diamond as target alternatives appeared to be a bit easier to discriminate. In metacontrast the faster sensory processes that help to build up target representation carry out contour and edge processing and the slower processes help carrying out area filling-in (of brightness and color) (von der Heydt, 2004). Consequently, metacontrast functions obtained with subjects focusing on fine contour or edge processing and on fine-scale spatial detail have function minima at much shorter target-mask stimulus onset asynchronies (SOAs) whereas area brightness dependent target evaluations produce function minima at a distinctly longer SOAs (Bachmann, 2009; Breitmeyer et al., 2006). In one of these studies (Bachmann, 2009) metacontrast functions for overall target-area contrast evaluation and for target-area internal contour contrast evaluation departed in opposite directions with an increasing SOA similarly to what was found by Albrecht et al. (2010). Thus, one objective for the second experiment would be to see whether the two types of subject populations have adopted different strategies for what to attend to in the target-mask combined images: typeA subjects may tend to discriminate fine detail of edges and base their reports on fine contours, type-B subjects may try to capture the overall filled-in shape of the whole target stimulus. It is also important to ascertain whether the two types of subjects populations is a general feature of metacontrast masking or is it related to specific varieties of metacontrast displays. For example, we should make sure whether these two types of populations emerge when smaller stimuli with no space between target (outer) and mask (inner) edges would be used (a third experiment). In Fig. 3 from Albrecht et al. (2010), we see that type-B observers tend to be slower in reacting to congruently primed masks compared to type-A observers. Thus, perhaps their perceptual speed is slower and at the instant when they become ready for evaluative analysis of what was briefly flashed, their phenomenal experience is inevitably based on different cues than those of type-A observers. 3. Is it a difference in pre-stimulus state of the brain? Performance in perceptual tasks involving variability of target-stimulus awareness depends not only on stimulation parameters and peculiarities of post-stimulus processing, but also on the pre-stimulus states. Recent evidence has shown this convincingly (Busch, Dubois, & VanRullen, 2009; Doesburg, Roggeveen, Kitajo, & Ward, 2008; Lakatos et al., 2009; Stokes, Thompson, Nobre, & Duncan, 2009; Womelsdorf, Fries, Mitra, & Desimone, 2006), including masking-studies (Aru & Bachmann, 2009; Hanslmayr et al., 2007; Lester et al., 1979; Mathewson, Gratton, Fabiani, Beck, & Ro, 2009; Rolke, 2008; Rolke,
T. Bachmann / Consciousness and Cognition 19 (2010) 667–671
669
Kleinmann, & Bausenhart, 2007). It follows that the two types of metacontrast functions may be a result of the spontaneous tendency or acquired skills of subjects to have different pre-stimulus states. For instance, pre-stimulus facilitatory effects may result from incresed gamma-band power (Aru & Bachmann, 2009), decreased alpha-band power (Hanslmayr et al., 2007; Womelsdorf et al., 2006), temporal coincidence of target presentation with the negative-trough of the slower EEG oscillations (see Whittingstall and Logothetis (2009), for delta-band involvement and Mathewson, Fabiani, Gratton, Beck, and Lleras (2010) and Osipova, Hermes, and Jensen (2008), for alpha-band involvement), basal forebrain activation (Goard & Dan, 2009), induction of cortical up-states (Castro-Alamancos, 2009) and/or resetting the phase of neural oscillations appropriately, particularly in the beta- and gamma-band ranges (Busch et al., 2009; Doesburg, Kitajo, & Ward, 2005; Hanslmayr et al., 2007; Mathewson et al., 2010). Thus, the fourth experiment would be to measure pre-stimuli states comparatively between type-B group and type-A group subjects (e.g., EEG or MEG can be comfortably used). Pre-stimulus signs of attentional activation or arousal (or preparatory retouch – Aru & Bachmann, 2009; Bachmann, 1994, 2007) may be more conspicuous in type-A subjects compared to type-B subjects. Because type-A subjects reach asymptotic levels of high performance already with a small increase in SOA, it is enough for them to have just a slight increase in mask-free time for the target so as to perceive it well. Thus, when pre-stimulus arousal, up-states or retouch processes are better prepared in time for the instant when target appears, target perception is more conspicuous. Purposeful manipulation of pre-stimulus states have been rarely done compared to the prevailing correlational studies. Yet, for example, non-specific thalamus of the patients was therapeutically activated by implanted electrodes immediately before the masking experiment (Bachmann, 1994). Instead of the typical expected type-B function of first stimulus identification, the masking function resembled type-A masking. A follow-up study showed that typical type-B functions emerge with Parkinsonian patients undergoing standard medication without direct thalamic brain stimulation (Bachmann et al., 1998). Therefore, the non-specific thalamic stimulation must have been the cause of the type-A effect. It may be that in Albrecht et al. (2010) type-A and type-B subjects may had differences in how thalamic modulatory processes participate in masking, which should be interesting to test. 4. Is learning to perform a metacontrast task specific to the particular stimuli displays and can different modes of metacontrast performance be taught? The results by Albrecht et al. (2010) were obtained just by one variety of the target and mask stimuli. In order to ascertain whether the two groups of subjects differentiated according to their qualitatively different masking functions can be found universally or not, varied types of metacontrast targets and masks should be used (e.g., in a fifth experiment). This is important also because recently metacontrast experiments where practice effects were studied lead only to type-A functions, although practice effects on metacontrast were found (Schwiedrzik, Singer, & Melloni, 2009). Because their stimuli and subject populations were different from those of Albrecht et al. (2010) and the number of subjects used in their published paper was not sufficiently large, the absence of bifurcation of metacontrast functions with increasing SOA need not mean that the results obtained by Albrecht et al. (2010) are doubtful. Moreover, when we analyze how target and mask stimuli that were used in the two studies look like we notice that stimuli in Albrecht et al. (2010) are closer to traditional metacontrast stimuli: solid dark area enclosed into a shape that fits closely into the inner aperture of the thick dark solid-line region surrounding the enclosed target. The stimuli used by Schwiedrzik et al. (2009), although formally satisfying the criteria that allow them to classify as metacontrast stimuli (no spatial overlap between target and mask edge contours), when taken from the ‘‘commonsense” point of view appear subjectively more like pattern- or form-masking stimuli. If we look at the thin-lined star-drawings of target and mask from Schwiedrzik et al. (2009) and flash them successively, we can not escape impression of two forms made up from thin lines overlapping spatially and crossing over each other. The metcaontrast aspect of the spatial arrangement of targets and masks applies only to very fine spatial scale of observation, but from the point of view of global scale of spatial arrangement of the stimuli when small local detail is ignored, the stimuli appear as the ones where their form-defining lines robustly cross over each other. In Albrecht et al. (2010), all of what makes up the target is located inside the mask contours; in Schwiedrzik et al. (2009), target aspects remaining inside the mask inner edge are in turn flanked by the mask aspects remaining inside the target. Therefore, the stimuli in Schwiedrzik et al. (2009) may tap pattern-masking effects more than metacontrast-masking effects. The absence of type-B functions is not surprising. Apart from the possible stimuli-dependence of the two types of expression of metacontrast, a general learning related question can be asked: is it possible to teach subjects of one group to develop masking functions of the other group and vice versa. Or would the learning-dependent transfer to the non-typical metacontrast function be symmetrical between the two groups or would this kind of transfer be possible with only one of the types? Thus, a sixth experiment should try test this. 5. Methodological implications The importance of the work by Albrecht et al. (2010) can be summarized by pointing out the basic methodological and research-strategy related issues emerging due to this article. First, there is the issue of universality of masking models. The search for an universal model of masking may remain utopical. Although a lot of attempts to model masking include dependency on stimulation parameters (Bachmann, 1994; Breitmeyer et al., 2006; Breitmeyer & Ög˘men, 2006; Dombrowe, Hermens, Francis, & Herzog, 2009; Duangudom, Francis, & Herzog, 2007; Enns, 2004), masking as a phenomenon may be
670
T. Bachmann / Consciousness and Cognition 19 (2010) 667–671
much more related to individual differences in the learned perceptual skills and decision-level strategies (biases) and originate not so much from the workings of hardwired sensory-perceptual or bottom-up attentional mechanisms. Modeling masking should benefit considerably from the future data about what makes the basis of individual differences in the qualitative expression of metacontrast masking. Second, there is the issue of the origin of differences – they may result from individual neurobiological differences or from learned behavioral differences. A metacontrast study with monozygotic twins would be experiment number seven. Third, if masking is closely related to the mechanisms of perceptual awareness (as I do believe) then we should start thinking about the possibly real individual variability of the visual awareness mechanisms. Thereby, the questions about inherited versus learned sensory-perceptual processes necessary for conscious-level experience will emerge. We may learn to be perceptually conscious and do this somewhat differently between ourselves. Fourth, masking is massively used as a tool in consciousness research and in most cases researchers implicitly assume that masking effects follow always the same universal rules of dependency on spatiotemporal factors, including SOAs. Moreover, many of these studies use pools of subjects that are small in number (typically between 4 and 10). The fact that masking may influence target perception in dramatically different ways depending on who is the subject in your experiment means that using masking as a method for controlling consciousness of targets or varying qualitative differences in the appearance of targets can hopelessly mess up your research. Therefore, stimulation parameters that help to avoid multiplicity of masking types and sufficient numbers of subjects should be used. Importantly, before the main experiment with using masking as a tool it may be good to ascertain if there is any qualitative variance in types of masking with your subjects. Sufficient idiographic care may become beneficial for a more sound nomothetic generalizations. Acknowledgments Thanks go to Estonian Science Foundation as this work was partly supported by Grant #7118 and also to Carolina Murd who prepared the replicas of the stimuli used by Albrecht et al. and programmed the experiment for the small informal study we ran. References Albrecht, T., Klapötke, S., & Mattler, U. (2010). Individual differences in metacontrast masking are enhanced by perceptual learning. Consciousness and Cognition, 19, 656–666. Aru, J., & Bachmann, T. (2009). Occipital EEG correlates of conscious awareness when subjective target shine-through and effective masking are compared: bifocal early increase in gamma power and speed-up of P1. Brain Research, 1271, 60–73. Bachmann, T. (1994). Psychophysiology of visual masking: The fine structure of conscious experience. Commack, NY: Nova Science Publishers. Bachmann, T. (2000). Microgenetic approach to the conscious mind. Amsterdam/Philadelphia: John Benjamins. Bachmann, T. (2007). Binding binding: Departure points for a different version of the perceptual retouch theory. Advances in Cognitive Psychology, 3, 41–55. Bachmann, T. (2009). Metacontrast masking of target-area internal contours and target overall surface brightness: The case of mutually coherent and incoherent visual objects. Spatial Vision, 22, 127–146. Bachmann, T., Asser, T., Sarv, M., Taba, P., Lausvee, E., Põder, E., Kahusk, N., & Reitsnik, T. (1998). Speed of elementary visual recognition operations in Parkinson’s disease as measured by the mutual masking method. Journal of Clinical and Experimental Neuropsychology, 20, 118–134. Breitmeyer, B. G., Kafalıgönül, H., Ög˘men, H., Mardon, L., Todd, S., & Ziegler, R. (2006). Meta- and paracontrast reveal differences between contour- and brightness-processing mechanisms. Vision Research, 46(17), 2645–2658. Breitmeyer, B. G., & Ög˘men, H. (2006). Visual masking: Time slices through conscious and unconscious vision. Oxford: Oxford University Press. Busch, N., Dubois, J., & VanRullen, R. (2009). The phase of ongoing EEG oscillations predicts visual perception. The Journal of Neuroscience, 29, 7869–7876. Castro-Alamancos, M. A. (2009). Cortical up and activated states: Implications for sensory information processing. The Neuroscientist, 15, 625–634. Doesburg, S. M., Kitajo, K., & Ward, L. M. (2005). Increased gamma-band synchrony precedes switching of conscious perceptual objects in binocular rivalry. NeuroReport, 16, 1139–1142. Doesburg, S. M., Roggeveen, A. B., Kitajo, K., & Ward, L. M. (2008). Large-scale gamma-band phase synchronization and selective attention. Cerebral Cortex, 18, 386–396. Dombrowe, I., Hermens, F., Francis, G., & Herzog, M. H. (2009). The roles of mask luminance and perceptual grouping in visual backward masking. Journal of Vision, 9(11), 22. doi:10.1167/9.11.22. 1–11 . Duangudom, V., Francis, G., & Herzog, M. H. (2007). What is the strength of a mask in visual metacontrast masking? Journal of Vision, 7(1), 7. doi:10.1167/ 7.1.7. 1–10 . Enns, J. T. (2004). Object substitution and its relation to other forms of visual masking. Vision Research, 44, 1321–1331. Goard, M., & Dan, Y. (2009). Basal forebrain activation enhances cortical coding of natural scenes. Nature Neuroscience, 12, 1440–1445. Hanslmayr, S., Aslan, A., Staudigl, T., Klimesch, W., Herrmann, C. S., & Bäuml, K.-H. (2007). Prestimulus oscillations predict visual perception performance between and within subjects. NeuroImage, 37, 1465–1473. Lakatos, P., O’Connell, M. N., Barczak, A., Mills, A., Javitt, D. G., & Schroeder, C. E. (2009). The leading sense: Supramodal control of neurophysiological context by attention. Neuron, 64, 419–430. Lester, M. L., Kitzman, M. J., Karmel, B. Z., Crow, G. J., Giambalvo, V., & Sidman, R. D. (1979). Neurophysiological correlates of central masking. In H. Begleiter (Ed.), Evoked brain potentials and behavior (pp. 525–544). New York: Plenum. Mathewson, K. E., Fabiani, M., Gratton, G., Beck, D. M., & Lleras, A. (2010). Rescuing stimuli from invisibility: Inducing a momentary release from visual masking with pre-target entrainment. Cognition. Mathewson, K. E., Gratton, G., Fabiani, M., Beck, D. M., & Ro, T. (2009). To see or not to see: Prestimulus alpha phase predicts visual awareness. The Journal of Neuroscience, 29, 2725–2732. Osipova, D., Hermes, D., & Jensen, O. (2008). Gamma power is phase-locked to posterior alpha activity. PLoS ONE, 3(12), e3990. doi:10.1371/ journal.pone.0003990. Rolke, B. (2008). Temporal preparation facilitates perceptual identification of letters. Perception & Psychophysics, 70, 1305–1313. Rolke, B., Kleinmann, K., & Bausenhart, K. M. (2007). Temporal preparation decreases metacontrast masking (Vol. 10). Tübingen: Tübinger Wahrnehmungskonferenz (TWK). Schwiedrzik, C. M., Singer, W., & Melloni, L. (2009). Sensitivity and perceptual awareness increase with practice in metacontrast masking. Journal of Vision, 9(10), 18. doi:10.1167/9.10.18. 1–18 .
T. Bachmann / Consciousness and Cognition 19 (2010) 667–671
671
Stokes, M., Thompson, R., Nobre, A. C., & Duncan, J. (2009). Shape-specific preparatory activity mediates attention to targets in human visual cortex. Proceedings of the National Academy of Sciences USA, 106, 19569–19574. von der Heydt, R. (2004). Image parsing mechanisms of the visual cortex. In L. M. Chalupa & J. S. Werner (Eds.), The visual neurosciences (pp. 1139–1150). MIT Press. Whittingstall, K., & Logothetis, N. K. (2009). Frequency-band coupling in surface EEG reflects spiking activity in monkey visual cortex. Neuron, 64, 281–289. Windelband, W. (1894). Geschichte und Naturwissenschaft. Straßburger Rektoratsrede. In W. Windelband (Ed.), Präludien. Aufsätze und Reden zur Philosophie und ihrer Geschichte (pp. 136–160). Tübingen: J.C.B. Mohr. Womelsdorf, T., Fries, P., Mitra, P. P., & Desimone, R. (2006). Gamma-band synchronization in visual cortex predicts speed of change detection. Nature, 439, 733–736.
Consciousness and Cognition 19 (2010) 672–673
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Reply
Individual differences in metacontrast masking: A call for caution when interpreting group data q Thorsten Albrecht *, Uwe Mattler Georg-Elias-Müller Institute for Psychology, Georg-August University, Göttingen, Germany
a r t i c l e
i n f o
Article history: Received 19 February 2010 Available online 3 April 2010 Keywords: Indiviudal differences Metacontrast masking Perceptual learning Consciousness
a b s t r a c t In this issue of Consciousness and Cognition, Bachmann (2010) comments on our study (Albrecht, Klapötke, & Mattler, 2010), which revealed two groups of observers with qualitative individual differences in metacontrast masking that are enhanced by perceptual learning. We are pleased that our study receives this attention and even more about Bachmann’s extremely positive comments. In this invited reply we argue that observers seem to be similar only at the beginning of the experiment but they have no choice as to which group to join. Findings strongly recommend to look at the data of individual subjects. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction In his commentary, Bachmann shares a variety of experimental ideas with the scientific community, which were stimulated by our study. Actually, several of the suggested experiments are currently run in our lab. Given the formal limitations of this reply and the preliminary status of our analyses, however, we decided to submit this work in a well prepared paper in the near future instead of giving a rush diminished report at this place. Nevertheless, we want to address three issues raised by Bachmann’s commentary in the following. 2. Initial similarities and differences in type-A and type-B observers In Bachmann’s perspective one specific feature of our study seems to be of particular importance. In our study, performance was essentially the same for type-A and type-B observers with the shortest stimulus onset asynchrony (SOA). Bachmann (2010) suggests that this supports the possibility that with short SOA the sensory contents of the perceptual images of targets were of the same type and of comparable quality for both groups. We did not address this feature in our article, although – as Bachmann pointed out – this feature is of theoretical importance because it could help to locate the source of individual differences between the two groups of observers. However, a preliminary analysis of data of a recent experiment with identical stimulation parameters, which we currently prepare for publication, revealed somewhat different results. At the end of one session of training, with the shortest SOA performance was worse for type-A observers (66.1%) than for type-B observers (87.8%). Initially, however, both groups performed similarly with this SOA (61.9% vs. 63.1%, respectively). Following Bachmann’s perspective, this finding suggests that all participants begin the experiment with the same
q Reply to Commentary by Bachmann, T. (2010). Individual differences in metacontrast: An impetus for clearly specified new research objectives in studying masking and perceptual awareness? Consciousness and Cognition, 19, 667–671. * Corresponding author at: Georg-Elias-Müller Institute for Psychology, Georg-August University Göttingen, Gosslerstr. 14, D-37073 Göttingen, Germany. Fax: +49 551 393662. E-mail address: [email protected] (T. Albrecht).
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.03.010
T. Albrecht, U. Mattler / Consciousness and Cognition 19 (2010) 672–673
673
sensory content at the shortest SOA. Further research is needed to determine communalities and differences in perceptual processing of type-A and type-B observers to locate the source of these individual differences. 3. A difference in direct phenomenal experience or in criteria? Bachmann (2010) highlights the question whether the two groups result because they experience the stimulus sequence in a different way or because participants choose to attend to different features of the stimulus sequence. As noted in our article, preliminary data of subjective reports, suggest that the phenomenological experience differs between the two groups of observers. Moreover, even when directly asked, type-A observers did not perceive features that were spontaneously reported by type-B observers, and vice versa. Nonetheless, it is unclear whether participants experience features of the stimulus sequence because they use them, or whether they use these features because they do not experience anything else. Beside this, however, two arguments suggest that participants are not able to choose which feature to use. First, if participants would have a choice between features it remains unclear why they do not use the feedback given in each trial to choose the optimal feature at short and long SOAs, respectively. Second, Experiment 4 of our study showed that not a single participant changed the group when we presented the target-mask sequence with a constant SOA for 288 trials. Moreover, following the idea that consciousness underlies neural plasticity (Cleeremans, 2007) the even more important question regarding the relation between consciousness and perceptual learning remains open for future research. The present data suggests that perceptual learning either improves conscious perception or the use of criteria. We think individual differences in metacontrast masking could become a fruitful tool to shed more light on this kind of questions. 4. The importance of the data of individuals Extending Bachmann’s (2010) methodological analysis, we would like to emphasize the importance of individual’s data. Since the beginnings of psychology as an experimental discipline in the latter half of the 19th century researchers like Wundt, Weber, Fechner and Ebbinghaus pursued the nomothetic approach aimed at finding general regularities and lawful expression of the effects of variables on conscious experience and behavior common to all or most of the individuals studied. Nonetheless, the founders of experimental psychology examined individual participants, but had to focus on the effects of powerful variables which could be detected easily on the background of irreducible variability. With the advent of inferential statistics, effects of more complex variables could be investigated because the weak and individually different effects of these variables could be averaged out by an increased number of participants. In consequence, experimental psychology tended to ascribe average effects of groups to individuals. Periodically researchers have advocated for the consideration of individual differences in experimental psychology (Cohen, 1994; Cronbach, 1957; Underwood, 1975). However, individual differences have not become a major issue in experimental psychology although there might be some renewed interest in recent years (e.g., Martens, Munneke, Smid, & Johnson, 2006; Schankin, Hagemann, & Wascher, 2009; Unsworth & Engle, 2007). The majority of these studies examines individual differences with a quantitative approach based on the correlation between dependent variables of individual participants. Along this line individual differences can be used to refine theoretical accounts. In stark contrast, the role of individual differences is fundamentally different in the study of Albrecht and colleagues (2010) because two groups of participants perform in qualitatively different ways in the same experimental situation. As elaborated by Bachmann (2010), this finding shows that the consideration of individuals’ data is crucial for metacontrast masking because it indicates that we may need two different theories. Beyond this, however, already the possibility of qualitative differences between individuals should alarm researchers in psychology to look at the data of individual participants. Moreover, the consideration of individuals’ data can provide additional benefits for theoretical work if there are phenomena which differ between individuals and other phenomena which are invariant. This has been demonstrated in our study which found invariant response priming functions in two groups with different masking functions (Albrecht et al., 2010, Experiments 1 and 2). Such a pattern of results indicates that at least one additional mechanism has to be assumed to account for individual differences. To sum up, we think our findings demonstrate that it is generally important for experimental psychologists to look at the data of individual subjects. References Albrecht, T., Klapötke, S., & Mattler, U. ( (2010). Individual differences in metacontrast masking are enhanced by perceptual learning. Consciousness and Cognition, 19, 656–666. Bachmann, T. (2010). Individual differences in metacontrast: An impetus for clearly specified new research objectives in studying masking and perceptual awareness? Consciousness and Cognition, 19, 667–671. Cleeremans, A. (2007). Consciousness: The radical plasticity thesis. In R. Banerjee & B. Chakrabarti (Eds.), Models of brain and mind: Physical, computational and psychological approaches (pp. 19–33). Elsevier. Cohen, R. L. (1994). Some thoughts on individual differences and theory construction. Intelligence, 18, 3–13. Cronbach, L. J. (1957). The two disciplines of scientific psychology. American Psychologist, 12, 671–684. Martens, S., Munneke, J., Smid, H., & Johnson, A. (2006). Quick minds don’t blink: Electrophysiological correlates of individual differences in attentional selection. Journal of Cognitive Neuroscience, 18, 1423–1438. Schankin, A., Hagemann, D., & Wascher, E. (2009). Inter-individual differences in change blindness. Psychophysiology, 46(Suppl.), S119. Underwood, B. J. (1975). Individual differences as a crucible in theory construction. American Psychologist, 30, 128–134. Unsworth, N., & Engle, R. W. (2007). The nature of individual differences in working memory capacity: Active maintenance in primary memory and controlled search from secondary memory. Psychological Review, 114, 104–132.
Consciousness and Cognition 19 (2010) 674–681
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Gambling on the unconscious: A comparison of wagering and confidence ratings as measures of awareness in an artificial grammar task q Zoltán Dienes a,*, Anil Seth b a b
Department of Psychology, University of Sussex, Brighton BN1 9QG, UK Department of Informatics, University of Sussex, Brighton BN1 9QJ, UK
a r t i c l e
i n f o
Article history: Received 27 January 2009 Available online 13 October 2009 Keywords: Unconscious knowledge Subjective measures Wagering Confidence Subjective threshold Artificial grammar learning Implicit learning Consciousness Higher order thoughts
a b s t r a c t We explore three methods for measuring the conscious status of knowledge using the artificial grammar learning paradigm. We show wagering is no more sensitive to conscious knowledge than simple verbal confidence reports but is affected by risk aversion. When people wager rather than give verbal confidence they are less ready to indicate high confidence. We introduce a ‘‘no-loss gambling” method which is insensitive to risk aversion. We show that when people are just as ready to bet on a genuine random process as their own classification decisions, their classifications are still above baseline, indicating knowledge participants are not aware of having. Our results have methodological implications for any study investigating whether people are aware of knowing. Ó 2009 Elsevier Inc. All rights reserved.
How can we tell if a person is aware of being in a mental state? For example, how can we tell if a person is aware of knowing? This is a crucial question for researchers interested in consciousness, and has been a core question in the fields of subliminal perception and implicit learning (Seth, Dienes, Cleeremans, Overgaard, & Pessoa, 2008). For example, Merikle (2007) defined subliminal perception as seeing without being aware of seeing, and Dienes (2008a) argued that implicit learning was a mechanism that produces knowledge one is not aware of. Further, if one accepts the proposal that unconscious mental states are mental states one is not aware of being in (Rosenthal, 2005), then any methodology for determining the conscious status of knowledge should measure as directly as possible whether or not the person is aware of knowing. Even if one does not subscribe to such ‘higher-order’ theories of consciousness (e.g. Block, 2001), determining when a person is aware of knowing still remains of theoretical and applied importance. Measuring awareness of being in a mental state is therefore a key issue whatever a priori assumptions about consciousness one holds (Seth et al., 2008). On the face of it, the most direct way of measuring awareness of being in a mental state is to ask a person whether they are in that mental state. Just so, verbal confidence ratings ask the person to indicate whether they are guessing or know to some degree. When a person says that they are literally guessing, that they know nothing, then the person is, on the face of it, not aware of knowing. In fact, when people say they are literally guessing, they can still discriminate stimuli at above chance levels in certain perceptual tasks (e.g. Weiskrantz, 1997) and they can still discriminate whether letter and other sequences have a certain structure in implicit learning paradigms (e.g. Dienes & Altmann, 1997; for a review see Dienes, 2008a). These results satisfy the guessing criterion of unconscious knowledge (Dienes, Altmann, Kwan, & Goode, 1995): When the person
q
This article is part of a special issue of this journal on commentary invited. * Corresponding author. Fax: +44 1273 6785058. E-mail address: [email protected] (Z. Dienes).
1053-8100/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2009.09.009
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
675
claims to be guessing they have above baseline discrimination performance. Similarly, if a person cannot tell what mental state they are in, there should be no relation between confidence and accuracy: This is the zero correlation criterion of unconscious knowledge (Dienes et al., 1995), and has been applied to both subliminal perception (Kolb & Braun, 1995; Kunimoto, Miller, & Pashler, 2001) and implicit learning (e.g. Dienes & Longuet-Higgins, 2004; Johansson, 2009; Tunney, 2005; for a review see Dienes, 2008a). Verbal confidence reports may have face validity for human adults as a measure of awareness of knowing, but they are problematic for at least young children and non-human animals (e.g. Seth, Baars, & Edelman, 2005). Ruffman, Garnham, Import, and Connolly (2001) wished to apply the confidence measures methodology of Dienes et al. (1995) to the question of whether three-year olds might have an unconscious theory of mind before they have a conscious one. However, a three-year old may not properly understand what confidence terms mean. So Ruffman et al. asked children to gamble tokens on different choice alternatives, showing that children’s gambling could be sensitive to objective probabilities but not false belief task choices. The amount a child was willing to gamble (children had 10 tokens to distribute among the choices) was taken to be a measure of subjective probability or confidence. Thus, the results indicate a lack of awareness of knowing the correct answer by the zero correlation criterion (lack of positive relation between gambling and correctness of answer). Similarly, Shields, Smith, Guttmannova, and Washburn (2005), Son and Kornell (2005), and Kornell, Son, and Terrace (2007) used gambling to determine whether Rhesus monkeys were aware of knowing. They trained Rhesus monkeys to wager differing amounts of tokens on their perceptual and memory judgments, showing a relation between amount wagered and accuracy: This is conscious knowledge by the zero correlation criterion if we accept wagering as indicating awareness of knowing. With the same assumption, their data also indicated unconscious knowledge by the guessing criterion, in that when Rhesus monkeys wagered the lowest amount, they were still significantly and substantially above chance baseline. Shields et al. (2005) used the same wagering method with adult humans as they had used with Rhesus macaques to show analogous cross-species effects. But is there any advantage in using wagering over verbal confidence in showing awareness of knowing in adult humans with well developed linguistic abilities? Like Shields et al., Persaud et al. asked adult humans to wager either a small or large amount (e.g. one versus two tokens) on the correctness of a decision. If the decision was correct the participant won the amount they wagered; if the decision was incorrect, the participant lost the amount wagered. Unlike Shields et al., Persaud, McLeod, and Cowey (2007) argued that wagering constitutes a gold standard for measuring awareness (cf. also Koch & Preuschoff, 2007, and contrast Seth, 2008). We will address in this paper the question of whether such wagering constitutes the most sensitive measure we have of the conscious status of knowledge, that is, of whether people are aware of knowing. Is there a reason to expect wagering rather than verbal confidence as being more direct, intuitive or sensitive as a measure of the conscious status of knowledge, as Persaud et al. claim? Verbal confidence can be susceptible to bias: People may think they know to some degree but say they know nothing at all. Of course, people can also think they know to some degree, and only wager low. Conversely one can wager high as a guess. Yet, if a conviction in a belief is to be shown in any way, it may plausibly be shown by the amount one is willing to stake on it. And when an amount of money is thrown down on the table, surely it lays public one’s convictions not only to others but also to oneself. So gambling may often go hand-in-hand with awareness of one’s own convictions; and hence, if those convictions are reliably caused, of one’s states of knowing. Still, some forms of gambling may be better in this role than others (cf. Mellor, 1971, 1991). On what grounds can it be argued that specifically wagering measures awareness of knowing? The expected pay off from a wager is the subjective probability of being correct multiplied by the amount of the wager. Consequently, if a person is aware of knowing to any degree, they should always go for the highest wager. Thus, the use of a low wager implies the person thinks they know nothing. This seems a somewhat sophisticated train of reasoning and it is not clear people directly or intuitively grasp it. For example, Shields et al. (2005) asked people to wager one, two or three tokens, where one won or lost the amount wagered. In terms of expected pay off, if one had any confidence at all, one should always go for the highest wager. If one had no confidence at all, then from the person’s point of view, the amount wagered does not matter. However people were more accurate for medium than low wagers, indicating people were not wagering optimally (in terms of expected pay off of tokens) given their confidence. One explanation for suboptimal wagering performance is that wagering likely involves trial by trial considerations of risk (or loss) aversion. For example, the prospect of losing $10 is often considered more salient than the prospect of winning the same amount (Kahneman & Tversky, 1979). These considerations are extra to whether or not one is aware of knowing; one could for example wager low even though one had some confidence in order to minimise loss. On this argument, verbal confidence rather than wagering may be the more direct and intuitive measure of awareness of knowing. In the following experiments we use a simple measure of risk aversion to determine empirically whether or not confidence or wagering are sensitive to risk aversion. Persaud et al. used artificial grammar learning (Reber, 1967, 1989) as a task to produce knowledge that could be conscious or unconscious: Participants memorised strings of letters, which unbeknownst to participants were generated by a set of rules. Participants were then informed of the existence of rules, though not what they were, and were asked to classify new strings as obeying the rules or not. We already know that people perform at above baseline levels when they believe they are guessing (by their verbal report), indicating some unconscious knowledge by the guessing criterion, and typically (though not always) people also show a relation between confidence and accuracy, showing some conscious knowledge by the zero correlation criterion (e.g. Dienes et al., 1995). Persaud et al. showed the same conclusions follow with wagering: People perform at above chance levels when they use the low wager, indicating unconscious knowledge by the guessing cri-
676
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
terion, and they show higher accuracy for high rather low wagers, indicating some conscious knowledge by the zero correlation criterion. We use the artificial grammar learning paradigm in two experiments to compare the measurement properties of verbal confidence and gambling. In Experiment 1 we investigate the wagering used by Persaud et al. and in Experiment 2 we consider another gambling measure of confidence, taken from the literature on the philosophy of subjective probability (Dienes, 2008b; Hacking, 2001). 1. Experiment 1 Experiment 1 used an artificial grammar learning paradigm with two groups of participants: One group wagered high or low on their decision and the other gave a binary confidence judgment (‘guess’ vs. ‘sure’). When a person is aware of knowing, they know when they know and when they are guessing. That is, a good measure of awareness of knowing will correlate with accuracy if there is any awareness of knowing at all (see Dienes, 2004; Dienes & Perner, 2004, for assumptions). Thus, the more sensitive a measure of awareness of knowing is, the stronger will be the relationship between the measure and accuracy (e.g. Tunney & Shanks, 2003). The aim of Experiment 1 was to determine whether wagering has a stronger relation with accuracy than confidence does. In addition participants were given a test of risk aversion (Hartog, Ferrer-i-Carbonell, & Jonker, 2000); we predicted risk aversion would correlate with wagering as a measure of awareness, but not with verbal confidence as a measure of awareness.
2. Method 2.1. Participants Seventy University of Sussex students aged between 20 and 23 were assigned to one of the two groups such that the wagering group had 40 participants and the verbal confidence group 30. 2.2. Materials Participants were trained on one of two grammars in order to employ the two grammar design of Dienes and Altmann (1997). In the training phase, half the participants are were asked to memorise strings of letters generated by one grammar (grammar ‘A’) and the other half of the participants were asked to memorise strings of letters generated by another grammar (‘B’). The test set consisted of an equal mix of A and B strings so that the strings which were grammatical for half the participants were ungrammatical for the other half. Thus, the appropriate baseline performance in the test phase is 50%. The two grammars (A and B) and the exact training and test strings used were those used by Dienes and Scott (2005, Experiment 2). The grammars were originally used by Reber (1969). The elements of the grammars were the letters M, T, V, R, and X. The generated strings were between 5 and 9 letter in length. The training phase consisted of 15 strings presented three times in a different random order each time (the same random order for each participant). The test consisted of 30 new letter strings from each grammar. See Dienes and Scott (2005, Appendix A) for a complete listing of the materials. A fixed order of test items was used; half the participants received the test items in that order, and half in the reverse order. Microsoft PowerPoint was used to present both training and test strings. Each string was presented on a separate slide displayed centrally in black text (Times New Roman font size 40) on a white background. The PowerPoint presentation for the training phase displayed each string for 5 s followed by a blank screen for a further 5 s. The presentation for the testing phase allowed participants to advance through the strings at their own pace. 2.3. Procedure For the training phase, participants were required to memorise each string while it was displayed and to write down what they could remember while the screen was blank. For the test phase, participants were informed that the order of letters in the strings seen during the training phase had obeyed a complex set of rules and that exactly half of the strings they were about to see obeyed the same rules. For each string, participants were required to indicate whether or not it obeyed the rules, and then to wager or to give their confidence (depending on the group). Participants in the wagering group were told at the beginning of the test phase that they started with 10 sweets. Then on each trial they indicated whether they wished to wager one or two sweets (their choice of Smarties or Haribo). They were told that if they were correct they would gain the amount wagered, but if they were incorrect they would lose that amount. Participants did not learn what they had won or lost until the end of the test phase, at which point they were given their earnings. No participant had to pay any sweets if they ended in debt. Participants in the confidence group indicated on each trial whether they were guessing or sure to some degree; guessing was defined as having no knowledge whatsoever, their answer was as good as flipping a coin. At the end of the experiment the last 50 participants run were given a test of risk aversion (Hartog et al., 2000). Specifically they were asked: ‘‘If there was a lottery for a 10 £ prize, which will be given to one of the 10 ticket holders, how much would you pay for a ticket? The ‘mathematical’ answer might seem to be ‘one pound’ because you have a 1/10 chance to win 10 £. But there is no right answer because you might not wish to certainly give up one pound for the mere probability of
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
677
getting something back. So think about what you personally would really be willing to pay – it is purely down to personal preference.’ Then participants were asked ‘‘If the prize were 100 £, which will be given to one of the 10 ticket holders, how much would you pay for a ticket?” The smaller the number to either question is given, the more risk averse they are. The second question involves a greater element of risk. For simplicity, we summed the two answers together to get an overall measure of risk aversion. 3. Results
a was at .05 for all analyses. The overall percent correct classification was 65% (SE = 1.9%) for the wagering group and 67% (SE = 2.2%) for the verbal confidence group, which did not differ significantly, t(68) = 0.64, 95% CI [7.6, 3.9]. Each group differed significantly from a chance baseline of 50%, ts of 7.89 and 7.28 respectively. Thus, participants in both groups learnt to discriminate the two grammars to about the same degree. Fig. 1 shows the percent correct classification of grammaticality for low and high confidence for each group. Overall, collapsing across whether people wagered or gave verbal confidence, people classified more accurately when they had high (73%) rather than low (60%) confidence, t(68) = 7.57. This 13% difference in accuracy between high and low confidence we will call the slope measure of the confidence–accuracy relationship. The fact that slope is greater than zero indicates conscious knowledge by the zero correlation criterion. Crucially, the difference in slopes between wagering (10.5, SE = 2.18) and verbal confidence (15.2, SE = 2.63) was not significant, t(68) = 1.41, 95% CI [11.5%, 2.0%]. The confidence interval indicates that we can reject a slope for wagering that is larger than for verbal confidence by any more than 2%. That is, to a high degree of sensitivity, we can assert that wagering is not more sensitive than verbal confidence for measuring awareness of knowing. Both wagering and verbal confidence measures indicate unconscious knowledge by the guessing criterion, as shown in Fig. 1 (the 95% confidence interval on accuracy for just the low confidence responses excludes 50% for both wagering and verbal confidence). We also measured the confidence–accuracy relationship with Type II d0 , which takes a hit to be a high confidence response when the grammaticality classification was correct and a false alarm to be a high confidence response when the classification was incorrect. That is, Type II d0 measures the ability of the participant to distinguish states of correctness (and thus, different states of knowledge) by their confidence: i.e. it is another way of operationalising the zero correlation criterion. Logistic Type II d0 was 0.33 (SE = .07) for wagering and 0.46 (SE = .08) for verbal confidence, both significantly above zero, indicating conscious knowledge by both confidence measures according to the zero correlation criterion. Crucially the d0 s did not differ significantly, t(68) = 1.16, 95% CI [.61, .16]. The confidence interval shows that we can rule out wagering having a higher d0 than verbal confidence by any amount above 0.16. That is, to a high degree of sensitivity, we can assert that wagering is not more sensitive than verbal confidence for measuring awareness of knowing.
Fig. 1. Percent correct scores for Experiment 1.
678
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
The proportion of low confidence responses was higher for wagering (63%, SE = 2.0%) than for verbal confidence (46%, SE = 3.1), t(68) = 4.61, indicating people were more willing to express some confidence verbally than with wagering. As would be expected, slope and Type II d0 correlated highly, r = .94 (N = 70). Thus, on current evidence, it does not seem to matter whether the zero correlation criterion is operationalised as slope or d0 (see Dienes, 2008a for discussion of these and other measures of the zero correlation criterion). Evans and Azzopardi (2007) criticised Type II d0 measures because in their study they found it correlated with bias; in our study, neither slope nor Type II d0 correlated significantly with the proportion of low confidence responses, rs of .19 and .23, respectively. The optimal score on the risk aversion measure, showing no risk aversion, is 11. The lower the number below 11 the greater the risk aversion. The wagering group’s risk aversion was 8.3 (SE = 0.99) and the verbal confidence group’s was 9.6 (SE = 1.23). For the wagering group the correlation between risk aversion and Type II d0 was significant, r(N = 30) = .52, and between risk aversion and slope marginally so, r = .32, p = .091. That is, the more risk averse a person was, the lower the measured amount of conscious knowledge using wagering. On the other hand, when verbal confidence was used the correlations were non-significant, .28 and .30 respectively (N = 20). For Type II d0 the correlation was significantly higher for wagering than for verbal confidence, z = 2.79, and for slope the correlation was also significantly higher for wagering than for verbal confidence, z = 2.07. In the verbal confidence group, risk aversion did not detectably correlate either with the percentage correct when confidence was low (guessing criterion measure of unconscious knowledge), r = .22, 95% CI [.25, .60], nor with the percentage correct when confidence was high, r = .05, 95% CI [.48, .40]. In the wagering group, risk aversion also did not detectably correlate with percentage correct when wagering low, r = .28, 95% CI [.09, .58]; in contrast, risk aversion did correlate with percentage correct when wagering high, r = .50. That is, the higher the risk aversion the lower the accuracy when high wagers were used: High wagers may be precisely the time a high risk averse person is most anxious.
4. Discussion Experiment 1 indicated that any claim that wagering is more sensitive than verbal confidence as a general measure of awareness of knowing can be ruled out to a high degree of sensitivity (compare numerically similar results in Table 3 of Persaud & McLeod, 2008). In fact, people were substantially more likely to indicate some confidence with a verbal confidence response rather than with wagering. This is not surprising given that wagering, more so than verbal confidence, is in principle susceptible to the affects of risk aversion. A person can wager low in order to avoid large losses, regardless of whether they have some confidence in their answer. Indeed, we found a measure of risk aversion correlated with the measured amount of conscious knowledge when wagering was used, but not when verbal confidence was used. Empirically, wagering as a measure of the conscious status of knowledge depends on how risk or loss averse a person is. Persaud et al. (2007) used substantial real money wagers (£1 vs. £2) in two of their studies. It may be that real money increases motivation and hence the sensitivity of wagering to conscious knowledge. On the other hand, real money may increase the influence of risk aversion on wagering, decreasing the sensitivity of wagering to conscious knowledge. In their artificial grammar learning study with one or two pound wagers, the slope of the accuracy–confidence relation was 8%, numerically no greater than the 11% slope we found using sweets. That is, sweets are just as effective as substantial money wagers in discriminating states of knowing from guessing. Other studies in Persaud et al. used only token money. In contrast with token money, but like real money, sweets provide real losses and gains. Mellor (1971, 1991) argued that a good measure of the strength of conscious belief is not the amount wagered for fixed odds (as we did in Experiment 1) but by the odds for an unknown stake and direction of bet, because the latter reduces the impact of psychological factors that could affect wagering but do not reflect strength of conscious beliefs. Hacking (2001) and Dienes (2008b) illustrated subjective probabilities with a different method that is easier for people to intuitively grasp than Mellor’s yet also deals with the problems with wagering. Indeed, the method is really a way of making concrete the instructions given with verbal confidence measures. When we ask people to give a confidence rating it is important that they understand precisely what we want them to understand the different ratings to mean. The scale points of a 1–7 confidence scale labelled from ‘not very confident’ to ‘very confident’ are open to different interpretations by different participants. To determine if people know something when they believe they know nothing at all (the guessing criterion of unconscious knowledge), the confidence rating for ‘guessing’ must be clearly defined – not just ‘‘not very confident”, for example. We have standardly (e.g. Dienes & Scott, 2005; Dienes et al., 1995) made clear to participants ‘guessing’ means ‘‘you know nothing at all – you could just as well flipped a coin to determine your answer”. The natural extension of these instructions is to actually flip a coin, or use some other equivalent process that is transparently random, and see if participants are indifferent between betting on a coin flip and their grammaticality decision. The typical objection to verbal confidence is that people may use terms like ‘guessing’ in idiosyncratic ways (Reingold & Merikle, 1993). Indeed, in everyday life the term ‘guessing’ allows a range of feelings of confidence. Apparent evidence for unconscious knowledge with the guessing criterion may arise only because of trials in which ‘guessing’ was used liberally by the participant when they were in fact aware of some degree of conviction (see Dienes, 2008a for discussion). However, if a person chooses to bet on a transparently random process rather than their own decision, the argument that the participant is using ‘guessing’ in an idiosyncratic way loses its force. Participants show what they mean by putting their money where their mouth is.
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
679
5. Experiment 2 Experiment 2 was similar to Experiment 1 except that only one form of confidence was taken, which we call no-loss gambling. After each grammaticality classification, the participant shuffled two cards and chose one. One of the cards had a reward (a sweet) indicated on its back and the other card did not. The participant was then asked which choice did they wish to bet on: The grammaticality decision or the card. If they were correct on their choice, they would gain a sweet. If they were incorrect, they lost nothing. Thus, there is no question of risk aversion because the participant never has an opportunity to lose sweets. The meaning of the choices should also be transparent, unlike with wagering. When the participant chooses to bet on the card, they have had every opportunity to appreciate they are betting on a random process rather their own classification decision: The participant is saying they are guessing in a clear way. 6. Method 6.1. Participants Thirty University of Sussex students aged between 20 and 23 took part in the experiment; they were in fact run at the same time as the participants in Experiment 1. 6.2. Materials Same as Experiment 1. 6.3. Procedure The same for experiment one with the exception only one form of confidence was taken. After participants classified a letter string, and were also asked to choose between two face-down cards, one of which had ‘‘SWEET” and the other had ‘‘NO SWEET” written on the face. Participants were asked whether they would like to either stay with their decision on the letter string which if correct would win a sweet, or turn-over their chosen card in an attempt to win a sweet. If they felt it did not matter which option they chose, they were asked to choose the card option (contrast wagering where a complete lack of confidence allows the participant to be indifferent between wagers). The cards were then shuffled by the experimenter, and then by the participant. Note we ensured that both the classification and card tasks involved active choices by the participant to eliminate potential biases against passively determined outcomes. Participants received their winnings in sweets at the end of the experiment. 7. Results Overall classification performance was 71% (SE = 2.0%), significantly above 50%, t(29) = 10.45, indicating participants had learnt to discriminate the grammars. Classification performance was 58% (SE = 2.6%) when participants bet on the cards, indicating unconscious knowledge by the guessing criterion, t(29) = 3.23, when problems of bias have been minimised. Classification performance was 75% (SE = 2.3%) when people preferred to bet on their classification. So slope is 17% (SE = 3.1%), significantly above zero, t(29) = 5.38, indicating conscious knowledge by the zero correlation criterion. Slope was not significantly different from the verbal confidence group of Experiment 1 (17% vs. 15%), t(58) = 0.39, 95% CI [6.6, 9.8], but was marginally higher than the wagering group of Experiment 1 (17% vs. 10%), t(58) = 1.73, p = .089. That is, no-loss gambling is as sensitive as verbal confidence. The proportion of low confidence responses (i.e. bets on the cards) was 28% (SE = 1.9%), significantly different from the percentage of low responses both for verbal confidence in Experiment 1 (46%), t(58) = 5.00, and for the wagering group of Experiment 1 (63%), t(68) = 12.33. That is, people were most likely to indicate confidence with no-loss gambling compared to verbal confidence and wagering. Average risk aversion was 8.7 (SE = .123). Risk aversion did not correlate with slope, r(N = 30) = .04, 95% CI [.03, .11]. The 95% confidence interval indicates we can reject any correlation with risk aversion above .11, i.e. we can to a high degree of sensitivity rule out risk aversion influencing no-loss gambling as a measure of conscious knowledge. 8. Discussion Experiment two introduced a no-loss gambling task that makes the meaning of ‘guess’ transparent. Importantly, when bias in interpretation of the meaning of ‘guess’ was thus minimised, the guessing criterion still indicated that people can know without being aware of knowing. Further, the confidence–accuracy relationship was of the same magnitude as with verbal reports and was marginally significantly stronger than with wagering. The tendency to state high confidence was also especially high when gambling on cards compared to verbal confidence and wagering. Although an apparently ‘high confi-
680
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
dence’ response in no-loss wagering may simply indicate indifference between the random process and grammaticality judgments, the tendency to indicate more confidence in no-loss gambling may appeal to researchers shy of over-estimating the amount of unconscious knowledge. No-loss gambling was designed to minimise risk aversion which was confirmed empirically. In sum, no-loss gambling shows how gambling can successfully be used to indicate awareness of knowing, with implications for measuring awareness of knowing in children and non-human animals as well as adults.
9. General discussion People show by their ability to discriminate the strings from two grammars that they are, in some sense, aware of properties that distinguish the grammars (Dienes & Seth, 2010; Seth et al., 2008). That is, such first-order awareness is knowledge that allows adaptive discriminations of states in the world (just as the sight of a blindsight patient does). Depending on one’s theoretical perspective (Seth et al., 2008), such first-order awareness could constitute phenomenal or primary consciousness, at least when perception is involved (Block, 2001; Snodgrass, Bernat, & Shevrin, 2004; Snodgrass & Shevrin, 2006) or alternatively an unconscious mental state (Rosenthal, 2005). How can we test whether people are aware of this first-order knowledge? We contrasted three techniques for measuring awareness of knowing. We argued wagering is susceptible to psychological factors irrelevant to awareness of knowing (cf. Clifford, Arabzadeh, & Harris, 2008; Mellor, 1971, 1991; Schurger & Sher, 2008) and presented data that wagering was no more sensitive than verbal confidence reports as a measure of awareness of knowing. Yet wagering involved a more cautious use of high confidence, so may be less useful than verbal reports in convincing sceptics of the existence of knowledge a person is not aware of. On the other hand, a no-loss gambling method involved the least caution in using high confidence responses. When bias in the interpretation of ‘guess’ was minimised because people bet on an actual random process, people could still have knowledge, indicating unconscious knowledge by the guessing criterion. Of course, people’s understanding of random processes can be faulty, and no measurement method a priori proves itself worthy (Chang, 2004; Seth et al., 2008). It is only by behaving well in theoretically motivated ways that any measure proves itself (see Dienes, 2008a, for a relevant discussion of this point for subjective measures in implicit learning research). In order to replicate Persaud et al. (2007) closely, we used their binary wagering, and hence also used a binary confidence scale to make a fair comparison, and a binary choice in no-loss gambling. Whether or not a binary scale is most sensitive is a matter of debate (cf. Overgaard, Fehl, Mouridsen, Bergholt, & Cleeremans, 2008; Overgaard, Rote, Mouridsen, & Ramsøy, 2006; Sergent & Dehaene, 2004 in the domain of perception). While e.g. Dienes et al. (1995) and Dienes and Altmann (1997) used a continuous 50–100% verbal confidence scale, Tunney and Shanks (2003) and Tunney (2005) found that a binary confidence scale was more sensitive. Dienes (2008a) found little to distinguish a variety of different confidence scales in the confidence–accuracy correlations in artificial grammar learning. However, with continuous scales, performance does improve gradually as confidence increases indicating people can make finer discriminations in their states of knowledge than just a binary one, at least when overall performance is good. The no-loss gambling method could be extended to different degrees of confidence by the participant choosing between their grammaticality choice and different card combinations, e.g. two cards providing a sweet and one that does not (cf. Hacking, 2001). Nonetheless, the most interesting choice for most researchers interested in the unconscious will be the one we gave participants: Between a completely random 50:50 process and their grammaticality (or perceptual) judgment. Confidence and gambling can both be used to measure whether people have higher order thoughts about knowing, that is, whether they are aware of knowing. Not everyone accepts the intuition that conscious knowledge is knowledge one is conscious of (contrast e.g. Cleeremans, 2008; Dienes & Perner, 1999; Rosenthal, 2005, with Block, 2001; Seth, 2008; see Seth et al., 2008). Absent such acceptance, our results are still important for establishing the presence of reflexive, ‘higher-order’, meta-, or introspective-consciousness, although on this view neither gambling performance nor confidence ratings can be taken as necessary for conscious knowledge per se (Seth, 2008, see Seth et al., 2008). Moreover, while it may be natural to take a person saying they have knowledge when they do as sufficient for indicating conscious knowledge, the same may not apply to gambling. For example, it may be possible for animals (and perhaps humans) to learn implicitly to gamble, and it is certainly possible for simple machine learning algorithms to learn advantageous wagering without being conscious (Cleeremans, Timmermans, & Pasquali, 2007). Gambling nonetheless has a clear purpose for measuring awareness of knowing in children and non-human animals: The fact that one can claim to have evidence for both conscious and unconscious knowledge in young children and monkeys is remarkable, even though the necessity and sufficiency of this evidence rests on further assumptions. Whatever a priori position one adopts, the distinction between knowledge one is or is not aware of is an important divide in nature with applied and theoretical significance (Dienes & Seth, 2010). This paper has dealt solely with the conscious status of what Dienes and Scott (2005) called judgment knowledge, e.g. knowledge that a string is or is not grammatical (or, in subliminal perception experiments, e.g. the judgement that a certain word was flashed on the screen). Judgement knowledge can be conscious even as the knowledge of the structure of the training strings that allowed such judgements remains unconscious. Dienes and Scott (2005) present a method for determining the conscious status of structural knowledge (see also Dienes, 2008a). While no-loss gambling and verbal confidence behaved similarly in the current experiments in terms of insensitivity to risk aversion, the use of gambling has some advantages over verbal confidence. As Persaud et al. (2007) indicate, participants find experiments with gambling fun. Gambling also has an intuitive appeal to researchers and its use may encourage more
Z. Dienes, A. Seth / Consciousness and Cognition 19 (2010) 674–681
681
people to study consciousness. No-loss gambling can also in principle be used with children and non-human animals. We encourage researchers to use both verbal confidence measures where convenient and no-loss gambling where appropriate. References Block, N. (2001). Paradox and cross purposes in recent work on consciousness. Cognition, 79, 197–219. Chang, H. (2004). Inventing temperature: Measurement and scientific progress. Oxford University Press. Cleeremans, A. (2008). Consciousness: The radical plasticity thesis. In R. Banerjee & B. K. Chakrabarti (Eds.), Progress in brain science (Vol. 168, pp. 19–33). Cleeremans, A., Timmermans, B., & Pasquali, A. (2007). Consciousness and metarepresentation: A computational sketch. Neural Networks, 20, 1032–1039. Clifford, C. W., Arabzadeh, E., & Harris, J. A. (2008). Getting technical about awareness. Trends in Cognitive Sciences, 12, 54–58. Dienes, Z. (2004). Assumptions of subjective measures of unconscious mental states: Higher order thoughts and bias. Journal of Consciousness Studies, 11, 25–45. Dienes, Z. (2008a). Subjective measures of unconscious knowledge. Progress in Brain Research, 168, 49–64. Dienes, Z. (2008b). Understanding psychology as a science: An introduction to scientific and statistical inference. Palgrave Macmillan. Dienes, Z., & Altmann, G. (1997). Transfer of implicit knowledge across domains? How implicit and how abstract? In D. Berry (Ed.), How implicit is implicit learning? (pp 107–123). Oxford: Oxford University Press. Dienes, Z., Altmann, G., Kwan, L., & Goode, A. (1995). Unconscious knowledge of artificial grammars is applied strategically. Journal of Experimental Psychology: Learning, Memory, & Cognition, 21, 1322–1338. Dienes, Z., & Seth, A. (2010). The conscious and the unconscious. In G. Koob, R. F. Thompson, & M. Le Moal (Eds.), Encyclopedia of behavioral neuroscience. Elsevier. Dienes, Z., & Longuet-Higgins, H. C. (2004). Can musical transformations be implicitly learned? Cognitive Science, 28, 531–558. Dienes, Z., & Perner, J. (1999). A theory of implicit and explicit knowledge. Behavioural and Brain Sciences, 22, 735–755. Dienes, Z., & Perner, J. (2004). Assumptions of a subjective measure of consciousness: Three mappings. In R. Gennaro (Ed.), Higher order theories of consciousness (pp. 173–199). John Benjamins Publishers: Amsterdam. Dienes, Z., & Scott, R. (2005). Measuring unconscious knowledge: Distinguishing structural knowledge and judgment knowledge. Psychological Research, 69, 338–351. Evans, S., & Azzopardi, P. (2007). Evaluation of a ‘bias-free’ measure of awareness. Spatial Vision, 20, 61–77. Hacking, I. (2001). Probability and inductive logic. Cambridge University Press. Hartog, J., Ferrer-i-Carbonell, A., & Jonker, N. (2000). On a simple survey measure of individual risk aversion. CESifo working paper series, CESifo working paper no. CESifo GmbH. Johansson, T. (2009). In the fast lane toward structure in implicit learning: Non-analytic processing and fluency in artificial grammar learning. European Journal of Cognitive Psychology, 21, 129–160. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291. Koch, K., & Preuschoff, K. (2007). Betting the house on consciousness. Nature Neuroscience, 10, 140–141. Kolb, F. C., & Braun, J. (1995). Blindsight in normal observers. Nature, 377, 336–338. Kornell, N., Son, L. K., & Terrace, H. S. (2007). Transfer of metacognitive skills and hint seeking in monkeys. Psychological Science, 18, 64–71. Kunimoto, C., Miller, J., & Pashler, H. (2001). Confidence and accuracy of near-threshold discrimination responses. Consciousness and Cognition, 10, 294–340. Mellor, D. H. (1971). The matter of chance. Cambridge University Press. Mellor, D. H. (1991). Matters of metaphysics. Cambridge University Press. Merikle, P. (2007). Preconscious processing. In M. Velmans & S. Schneider (Eds.), The blackwell companion to consciousness (pp. 512–524). Blackwell. Overgaard, M., Fehl, K., Mouridsen, K., Bergholt, B., & Cleeremans, A. (2008). Seeing without seeing? Degraded conscious vision in a blindsight patient. PLoS ONE, 3(8), e3028. Overgaard, M., Rote, J., Mouridsen, K., & Ramsøy, T. Z. (2006). Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task. Consciousness and Cognition, 15, 700–708. Persaud, N., & McLeod, P. (2008). Wagering demonstrates subconscious processing in a binary exclusion task. Consciousness and Cognition, 17, 565–575. Persaud, N., McLeod, P., & Cowey, A. (2007). Post-decision wagering objectively measures awareness. Nature Neuroscience, 10, 257–261. Reber, A. S. (1967). Implicit learning of artificial grammars. Journal of Verbal Learning and Verbal Behavior, 6, 317–327. Reber, A. S. (1969). Transfer of syntactic structures in synthetic languages. Journal of Experimental Psychology, 81, 115–119. Reber, A. S. (1989). Implicit learning and tactic knowledge. Journal of Experimental Psychology: General, 118, 219–235. Reingold, E. M., & Merikle, P. M. (1993). Theory and measurement in the study of unconscious processes. In M. Davies & G. W. Humphreys (Eds.), Consciousness (pp. 40–57). Oxford: Blackwell. Rosenthal, D. M. (2005). Consciousness and mind. Oxford: Oxford University Press. Ruffman, T., Garnham, W., Import, A., & Connolly, D. (2001). Does eye gaze indicate implicit knowledge of false belief? Charting transitions in knowledge. Journal of Experimental Child Psychology, 80, 201–224. Schurger, A., & Sher, S. (2008). Awareness, loss aversion, and post-decision wagering. Trends in Cognitive Sciences, 12, 209–210. Sergent, C., & Dehaene, S. (2004). Is consciousness a gradual phenomenon? Evidence for an all-or-none bifurcation during the attentional blink. Psychological Science, 15, 720–729. Seth, A. K. (2008). Post-decision wagering measures metacognitive content, not sensory consciousness. Consciousness and Cognition, 17, 981–983. Seth, A. K., Baars, B. J., & Edelman, D. B. (2005). Criteria for consciousness in humans and other mammals. Consciousness and Cognition, 14, 119–139. Seth, A., Dienes, Z., Cleeremans, A., Overgaard, M., & Pessoa, L. (2008). Measuring consciousness: Relating behavioural and neurophysiological approaches. Trends in Cognitive Sciences, 12, 314–321. Shields, W. E., Smith, D. J., Guttmannova, K., & Washburn, D. A. (2005). Confidence judgments by humans and Rhesus monkeys. Journal of General Psychology, 213, 165–186. Snodgrass, M., Bernat, E., & Shevrin, H. (2004). Unconscious perception: A model-based approach to method and evidence. Perception and Psychophysics, 66, 846–867. Snodgrass, M., & Shevrin, H. (2006). Unconscious inhibition and facilitation at the objective detection threshold: Replicable and qualitatively different perceptual effects. Cognition, 101, 43–79. Son, L. K., & Kornell, N. (2005). Metaconfidence judgements in Rhesus macacques: Explicit vs. implicit mechanisms. In H. S. Terrace & J. Metcalfe (Eds.), The missing link in cognition: Origins of self-reflective consciousness (pp. 296–320). Oxford University Press. Tunney, R. J. (2005). Sources of confidence judgments in implicit cognition. Psychonomic Bulletin & review, 12, 367–373. Tunney, R. J., & Shanks, D. R. (2003). Does opposition logic provide evidence for conscious and unconscious processes in artificial grammar learning? Consciousness and Cognition, 12, 201–218. Weiskrantz, L. (1997). Consciousness lost and found. Oxford University Press.
Consciousness and Cognition 19 (2010) 682–684
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Commentary
Optimizing subjective measures of consciousness q Morten Overgaard a,*, Bert Timmermans b, Kristian Sandberg a, Axel Cleeremans b a b
Cognitive Neuroscience Research Unit, Hammel Neurorehabilitation and Research Center, Voldbyvej 15, 8450 Hammel, Denmark Consciousness, Cognition & Computation Group, Université Libre de Bruxelles, Brussels, Belgium
a r t i c l e
i n f o
Article history: Available online 25 January 2010 Keywords: Consciousness Methodology Introspection Visual consciousness Implicit learning
a b s t r a c t Dienes and Seth (2010) conclude that confidence ratings and post-decision wagering are two comparable and recommendable measures of conscious experience. In a recently submitted paper, we have however found that both methods are problematic and seem less suited to measure consciousness than a direct introspective measure. Here, we discuss the methodology and conclusions put forward by Dienes and Seth, and why we think the two experiments end up with so different recommendations. Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction The striking contemporary resurgence of interest in empirical approaches to consciousness has brought about a methodological challenge: What are we to make of subjective reports? Cognitive science, with a few exceptions (see e.g., ‘‘Verbal reports as data”, Ericsson & Simon, 1980) has tended to disregard them or to dismiss them altogether, preferring so-called objective data as the primary source of evidence regarding what a participant knows or does not know about a particular state of affairs. While this approach is legitimate in some cases, it becomes problematic when studying consciousness as conscious contents may dissociate from behavior. Subjective experience, indeed, cannot be observed ‘‘from the outside”. Therefore, in order to test the validity of any objective method, one is forced to somehow calibrate it using subjective methods, and thus a model for the use of subjective reports seems unavoidable. Nevertheless, despite a number of theoretical papers on subjective reports (e.g. Overgaard, Gallagher, & Ramsøy, 2008; Seth, Dienes, Cleeremans, Overgaard, & Pessoa, 2008), studies comparing objective and subjective methods have been few and in between. Dienes and Seth (2010), together with Sandberg, Timmermans, Overgaard, and Cleeremans (2010), are amongst the first to engage in this endeavor. In this commentary, we discuss Dienes and Seth’s study to evaluate the extent to which the experiments are suggestive of an ‘‘optimal” method to collect subjective reports, or whether this goal is complicated by confounding factors. In Dienes & Seth’s implicit learning paradigm, participants first memorized letter strings and were subsequently told that they had in fact been constructed according to specific ‘‘rules”. Participants were then to discriminate novel letter strings that did or did not obey these ‘‘rules”. One group of participants reported how confident they were about each decision (confidence rating (CR): ‘‘guessing” (no knowledge at all) or ‘‘sure to some degree”), and a second group had to wager (low or high; one or two sweets) on the correctness of each of their decisions (post-decision wagering (PDW)). Persaud and colleagues (Persaud & McLeod, 2008; Persaud, McLeod, & Cowey, 2007) have argued that PDW constitutes a better method
q Commentary on Dienes, Z., & Seth, A. (2010). Gambling on the unconscious: A comparison of wagering and confidence ratings as measures of awareness in an artificial grammar task. Consciousness and cognition, 19, 674–681. * Corresponding author. Fax: +45 89 49 44 00. E-mail address: [email protected] (M. Overgaard).
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2009.12.018
M. Overgaard et al. / Consciousness and Cognition 19 (2010) 682–684
683
to assess awareness, as the prospect of monetary gain motivates the participants to reveal all their knowledge. Thus, the measure should in theory minimize the risk of misclassifying conscious knowledge as unconscious. Contrary to this, however, Dienes and Seth found that CR is actually more sensitive than PDW – their wagering participants were reluctant to place high wagers due to risk aversion. In a second experiment, Dienes & Seth altered the way participants have to wager in two ways so as to minimize participants’ risk aversion (no-loss gambling). First, participants could wager only one sweet and no longer lost it when answering incorrectly, and second, instead of wagering on the correctness of their response, participants could now choose a random card draw with a 50% chance of obtaining a sweet. They were instructed to choose this option when, and only when they felt that their grammaticality decision was completely random. These instructions seem similar to the instructions for the CR in the first experiment in that the first response option is defined as ‘‘guessing”, and the second is defined as ‘‘anything that does not qualify as guessing”. Not surprisingly, the new instructions caused PDW to perform at the same level as CR. Based on their two experiments, Dienes and Seth conclude that CR and some form of PDW both are reasonable measures of consciousness that do not significantly vary from each other, and that both can be recommended for future studies. However, based on the above, it should be noted that PDW had to be substantially modified to produce results as good as CR. In this light, we raise a crucial issue: although Dienes and Seth point out the many similarities between awareness (and measures of awareness) in implicit learning and perceptual awareness paradigms, it is an empirical question whether Dienes & Seth’s conclusions about the measures (regarding their equivalence as well as their adequacy) extend to studies of perceptual awareness. We (Sandberg et al., 2010) recently conducted such a scale comparison experiment within the field of visual awareness. In the following we will discuss the consequences of our findings for the measures of awareness used in implicit learning paradigms. 2. Comparing measures of consciousness In our experiment, we (Sandberg et al., 2010) compared three measures of subjective perceptual experience: The Perceptual Awareness Scale (PAS), CR, and PDW. In the experiment, participants were briefly presented with one of four masked geometrical shapes and had to indicate the correct shape in a forced-choice task. Following each trial, participants had to indicate their subjective experience on a 4-point version of one of the three scales (each participant responded using only one scale), where only the instructions and the scale anchors differed (PAS: no experience, vague experience, almost clear experience, clear experience; CR: not confident at all, slightly confident, quite confident, very confident; PDW: imaginary amounts €5, €10, €15, €20). We determined (1) which scale correlated best and most consistently with performance (indicating awareness) and (2) whether we could detect above-chance performance in the absence of awareness and how the scales differ from each other in terms of revealing such unconscious processing. Regarding the extent to which reported consciousness correlated with performance, our results indicated that PAS was more exhaustive than CR in detecting conscious influences on performance, and that CR in turn was more exhaustive than PDW. Furthermore, as in Dienes & Seth’s first experiment, people using PDW were more inclined to give low wagers for stimulus intensities for which it was clear from PAS that stimuli were mostly perceived consciously. Since we did not include the Dienes & Seth suggested modifications to PDW, this result can be explained by participants’ risk aversion. However, and crucially, the same was true for CR: even though it fared slightly better than PDW, participants claimed to be guessing for stimuli that PAS showed to be processed, at least to some extent, consciously. As Dienes and Seth showed that risk aversion does not predict confidence ratings, the superior results of PAS over CR must be explained by something else. One straightforward yet somewhat controversial interpretation of our results is that participants perform the task exactly as instructed. In other words, when they are asked to specifically report what they experience, this is what they will do. Likewise, when asked to wager, they will perceive it as a gambling situation, and in this type of situation, factors such as emotional arousal, risk aversion, and gambling strategy may influence task performance and consciousness rating. Finally, when asked to report their confidence, participants consider how much they trust the correctness of the report they just issued. The crucial difference here lies in the fact that whereas PAS in principle refers to visual experience regardless of the issued response, both CR and PDW refer explicitly to so-called ‘‘judgment knowledge”, meaning that rather than being just a reflection of perceptual experience, participants have to evaluate how this knowledge contributed to their response. Consequentially and crucially, other cognitive processes than those specifically related to reporting the experience influence the results in these latter two cases. According to Dienes and Seth, ‘‘one could for example wager low even though one had some confidence in order to minimize loss”—similarly, one could have no confidence in the correctness of one’s response even though one had a partial or vague experience of the stimulus. Such a vague experience could for instance be seeing just a weak glimpse, or a part, of something on the screen, but having no idea what it was. On a PAS scale, a participant having this experience would report a ‘‘vague experience”, but in order to report some degree of confidence, the participant would have to know that this very vague experience is indeed enough to improve the accuracy of the identification task slightly. In other words, it may be a prerequisite for the use of a CR scale that subjects entertain the idea that even small changes in experience correlate with performance. Based on their two experiments, Dienes and Seth conclude that CR and some form of PDW both are reasonable measures of consciousness that do not significantly vary from each other, and that both can be recommended for future studies. However, in spite of a number of similarities between the Dienes and Seth study and our above-mentioned study (Sandberg et al.,
684
M. Overgaard et al. / Consciousness and Cognition 19 (2010) 682–684
2010), we arrive at different conclusions. Our study indicates problems that are shared by PDW and CR so that neither, rather than both, of them can be recommended to measure conscious experience in cases where it is possible to ask the subject to report their conscious experience directly. As our results based on the CR and PDW scales are highly similar to the findings of Dienes and Seth, it seems obvious to raise the question of whether our PAS findings can be extended to the implicit learning paradigm as well. 3. Different paradigms, different measures? Would a scale that does not require knowledge of the influence of particular aspects of experience on task performance also lead to superior results within an implicit learning paradigm? And is it at all possible to consctruct a PAS-like scale for an implicit learning paradigm? PAS ratings refer to the clarity of the perceptual experience. As such, the experimenter is simply asking the participants to report what they saw. In artificial grammar learning, the knowledge (‘‘structural knowledge”) that drives above-chance performance in the classification task may consist of abstract rules, of statistical regularities or of exemplars or fragments thereof (see Cleeremans et al., 1998; Cleeremans & Dienes, 2008). The corresponding conscious knowledge can be defined as involving awareness of what motivated your decision (judgment knowledge). From this perspective one could argue in favor of confidence ratings, or even wagering, because one is interested in whether people know on what knowledge they based their decision, without the experimenter probing one specific type of structural knowledge. However, the problem remains that participants fail to recognize what determined their performance, and claim to be guessing. Conversely, a high confidence rating need not reflect consciousness per se. Consider the following (very simple) example. One person who is a non-native English speaker knows that one says ‘‘I am” and not ‘‘I are”, and he is fully aware of why this is the case, whereas a native English speaker also knows that ‘‘I am” is correct, contrary to ‘‘I are”, but has no idea why this is the case. ‘‘I are”, to him, just sounds wrong. The two subjects may have the same amount of confidence that their report is correct, yet they are conscious in quite different ways. Before attempting to construct an awareness scale suited for the implicit learning paradigm, it would thus seem necessary to define the knowledge to be considered conscious knowledge, but this is no easy task. Confidence ratings have served as a convenient way of avoiding the definition of conscious knowledge, but with our recent empirical results there is reason to believe that this solution has flaws of its own. Thus, it seems to us that implicit learning researchers face a difficult dilemma. On the one hand, there is reason to believe that confidence ratings may not be the best way to gain knowledge about conscious experience, but on the other hand, it seems a non-trivial task to adjust a PAS-like scale to an implicit learning paradigm. This is because, if one would want to probe for conscious experience of structural knowledge, one faces two problems: first, what knowledge to probe; and second, how to probe for such knowledge without making this knowledge explicit to the participant? Nevertheless, it seems a logical conclusion, based on the above, that such practical difficulties should be overcome since current attempts to ‘‘objectify” measures of consciousness fare worse than does the direct, introspective measure. Acknowledgments Axel Cleeremans is a Research Director with the F.R.S.-FNRS (Belgium). This work was supported by The MindBridge project, funded by the European Commission under the Sixth Framework Programme, Contract No. 043457 and by Concerted Research Action 06/11-342 titled ‘‘Culturally modified organisms: What it means to be human in the age of culture”, financed by the Ministère de la Communauté Française – Direction Générale l’Enseignement non obligatoire et de la Recherche scientifique (Belgium). References Dienes, Z., & Seth, A. (2010). Gambling on the unconscious: A comparison of wagering and confidence ratings as measures of awareness in an artificial grammar task. Consciousness and cognition, 19, 674–681. Ericsson, K. A., & Simon, H. A. (1980). Verbal reports as data. Psychological Review, 87, 215–251. Overgaard, M., Gallagher, S., & Ramsøy, T. Z. (2008). Integration of first-person methodologies in cognitive science. Journal of Consciousness Studies, 15(5), 100–120. Persaud, N., & McLeod, P. (2008). Wagering demonstrates subconscious processing in a binary exclusion task. Consciousness and Cognition, 17(3), 565–575. Persaud, N., McLeod, P., & Cowey, A. (2007). Post-decision wagering objectively measures awareness. Nature Neuroscience, 10(2), 257–261. Sandberg, K., Timmermans, B., Overgaard, M., & Cleeremans, A. (2010). Measuring consciousness: Is one measure better than the other? Consciousness and cognition. doi:10.1016/j.concog.2009.12.013. Seth, A. K., Dienes, Z., Cleeremans, A., Overgaard, M., & Pessoa, L. (2008). Measuring consciousness: Relating behavioural and neurophysiological approaches. Trends in Cognitive Sciences, 12(8), 314–321.
Consciousness and Cognition 19 (2010) 685–686
Contents lists available at ScienceDirect
Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog
Reply
Subjective measures of implicit knowledge that go beyond confidence: Reply to Overgaard et al. q Zoltán Dienes a,c,*, Ryan B. Scott a,c, Anil K. Seth b,c a
School of Psychology, University of Sussex, Sussex BN1 9QH, United Kingdom School of Informatics, University of Sussex, Sussex BN1 9QG, United Kingdom c Sackler Centre for Consciousness Science, University of Sussex, Sussex BN1 9QG, United Kingdom b
a r t i c l e
i n f o
Article history: Received 26 January 2010 Available online 25 February 2010 Keywords: Subjective measures Confidence Implicit learning Unconscious knowledge
a b s t r a c t Overgaard, Timmermans, Sandberg, and Cleeremans (2010) ask if the conscious experience of people in implicit learning experiments can be explored more fully than just confidence ratings allow. We show that confidence ratings play a vital role in such experiments, but are indeed incomplete in themselves: in addition, use of structural knowledge attributions and ratings of fringe feelings like familiarity are important in characterizing the phenomenology of the application of implicit knowledge. Ó 2010 Elsevier Inc. All rights reserved.
Ramsøy and Overgaard (2004) introduced a scale of perceptual experience which has the goal of asking people about the clarity of contents of their visual experience, e.g. no experience, a glimpse, almost clear, clear. Sandberg, Timmermans, Overgaard, and Cleeremans (in press) and Overgaard et al. (2010) contrast this Perceptual Awareness Scale (PAS) with confidence ratings because, they argue, a person could in principle be aware of the clarity of visual contents independently of the person’s assessment of the relevance of the contents for a discrimination at hand. For example, one could clearly or vaguely consciously see something but think it is of no relevance to making a required discrimination, and so have no confidence in the discriminative response. In such a case the PAS scale would indicate some conscious visual experience of the stimulus even as confidence ratings indicated no confidence. Thus, Overgaard et al. argue that PAS is superior to confidence ratings for determining subliminal perception. They ask how the methodology in implicit learning could correspondingly be improved beyond confidence ratings to take into account the principles behind PAS. Dienes and Seth (in press) argue that in fact confidence ratings determine the conscious status of the visual contents relevant to a discrimination better than PAS does: Having some conscious content does not entail having relevant conscious contents. In the present comment, we recall that a methodology already exists in the implicit learning literature that determines conscious contents independently of a participant’s assessment of the relevance of those contents to a discrimination. We further argue that this method needs to be combined with confidence ratings for a more complete assessment of the conscious status of relevant mental states. Scott and Dienes (2008) used artificial grammar learning to explore the phenomenology of the application of knowledge resulting from implicit learning. In their Experiment 3, after classifying a test string, people reported their confidence in the classification, rated how familiar the string felt, and indicated what strategy they used to make the classification. The options for strategy use were: random response, intuition, familiarity, rules, or recollection. Importantly, people could respond with q Reply to Commentary by Overgaard, M., Timmermans, B., Sandberg, K., & Cleeremans, A. (2010). Optimizing subjective measures of consciousness. Consciousness & Cognition, 19, 682–684. * Corresponding author. Address: School of Psychology, University of Sussex, Sussex BN1 9QG, United Kingdom. E-mail address: [email protected] (Z. Dienes).
1053-8100/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2010.01.010
686
Z. Dienes et al. / Consciousness and Cognition 19 (2010) 685–686
no confidence and choose a strategy other than random responding; for example, if they were aware of using rules to make a judgment but had no confidence in those rules for determining grammaticality. Similarly, people could say they were responding randomly and still indicate what degree of familiarity the test string had, thereby reporting conscious experiences that may in fact be relevant, even if the person themselves had no idea about their relevance. As Sandberg et al. (in press) reported in the visual case, Scott and Dienes (2008) found dissociations between confidence in a decision and conscious experiences concerning the target of the decision. People could be aware of using rules, recollections or feelings of familiarity in classifying a string while having no confidence in that classification. In addition, like Sandberg et al., when people said they had no relevant conscious experience at all – no experience of the stimulus (in the visual case) or no experience of their decision having any basis whatsoever (in the implicit learning case) – people still classified above chance. Further, through the use of different scales, i.e. reports of different conscious contents, Scott and Dienes were able to reveal subtleties not apparent when only one scale is used. They showed, for example, that people could believe they were responding randomly while they were actually using the information contained in their feelings of familiarity (Scott & Dienes, 2008). They also showed that, in more complex situations, when people believed they were responding randomly, even the information contained in feelings of familiarity did not account for classification accuracy (Scott & Dienes, 2010; Wan, Dienes, & Fu, 2008). Importantly, the incorporation of confidence ratings also allows us to go beyond the PAS scale alone, to distinguish conscious states that it cannot, as Dienes and Seth (in press) argued for the visual case. In the visual case, a person can clearly consciously see an object as having property x without at all consciously seeing it has property y, even if the person did indeed (unconsciously) see the object had property y. One cannot just say ‘‘the seeing is conscious”: The mental state that was conscious must be fixed by specifying its contents, because different contents imply different mental states. What contents are relevant to implicit learning? Dienes and Scott (2005) distinguished knowledge contents concerning the structure of grammatical strings in general (structural knowledge) and the knowledge content that a particular string is (or is not) grammatical (judgment knowledge). Other relevant contents are the string being familiar to some degree. One could be aware of believing the string is grammatical but not of believing that grammatical strings can only start with M or T; similarly one can be aware of a feeling of familiarity but not of believing the string is grammatical. When a person gives a confidence rating in a classification of a string being grammatical, such a response measures specifically whether the person is aware of believing the string is grammatical, i.e. the conscious status of judgment knowledge. Structural knowledge may or may not be conscious; and feelings of familiarity may or may not be conscious. Thus, one can fail to be aware that one believes the string is grammatical even as one is aware of a degree of familiarity with the string. Understanding such fine gradations in phenomenology is only possible with multiple subjective measures. Similarly in the visual case, interesting distinctions can occur in conscious phenomenology which can only be revealed by multiple subjective measures, including confidence. As Weiskrantz (1997) found, a person can have conscious judgement knowledge as revealed by having some confidence even in the absence of any visual experience as such (type I vs. type II blindsight). Overgaard, Fehl, Mouridsen, Bergholt, and Cleeremans (2008) provide evidence in favor of such a conclusion, assuming the subject ‘‘performed the task exactly as instructed”: A blindsight patient who was prone to say she had not consciously seen a stimulus under certain conditions, would say she was aware of a stimulus being there but had no idea what it was. In summary, the use of confidence ratings, structural knowledge attributions, and ratings of other fringe feelings like familiarity can provide a detailed assessment of the conscious phenomenology of participants in implicit learning experiments, in a similar way as the combination of PAS and confidence ratings can in perception experiments. References Dienes, Z., & Seth, A. K. (in press). Measuring any conscious content versus measuring the relevant conscious content: Comment on Sandberg et al. Consciousness & Cognition. Dienes, Z., & Scott, R. B. (2005). Measuring unconscious knowledge: Distinguishing structural knowledge and judgment knowledge. Psychological Research, 69(5–6), 338–351. Overgaard, M., Timmermans, B., Sandberg, K., & Cleeremans, A. (2010). Optimizing subjective measures of consciousness. Consciousness & Cognition, 19, 682–684. Overgaard, M., Fehl, K., Mouridsen, K., Bergholt, B., & Cleeremans, A. (2008). Seeing without seeing? Degraded conscious vision in a blindsight patient. PLoS ONE, 3(8), e3028. Ramsøy, T. Z., & Overgaard, M. (2004). Introspection and subliminal perception. Phenomenology and the Cognitive Sciences, 3, 1–23. Sandberg, K., Timmermans, B., Overgaard, M., & Cleeremans, C. (in press). Measuring consciousness: Is one measure better than the other? Consciousness & Cognition. Scott, R. B., & Dienes, Z. (2010). Knowledge applied to new domains: The unconscious succeeds where the conscious fails. Consciousness & Cognition, 19, 391–398. Scott, R. B., & Dienes, Z. (2008). The conscious, the unconscious, and familiarity. Journal of Experimental Psychology – Learning Memory and Cognition, 34(5), 1264–1288. Wan, L. L., Dienes, Z., & Fu, X. L. (2008). Intentional control based on familiarity in artificial grammar learning. Consciousness and Cognition, 17, 1209–1218. Weiskrantz, L. (1997). Consciousness lost and found. Oxford University Press.