Attention Synchrony
According to Berlyne (1971), an interesting stimulus is
one which elicits an inclination to respond, but this tendency is
inhibited or blocked in some manner. The inhibited response
increases arousal until a new cognitive response appropriate to
the stimulus is formed. Once it does, arousal drops and this
reduction in arousal is rewarding. Subjective interest in a film
ought to a function of the number and magnitude of response
inhibition.
Response inhibition and interestingness
A film or story will be interesting when it deviates
from an expected direction, but only if it does so in a
meaningful way. For instance, the film "Chaplin" becomes
interesting when the fate of the title character reverses. The
first minute of the film preview presents Chaplin in a positive
light, building audience empathy for him. Once this positive view
is firmly established so that the audience expects additional
positive revelations, negative scenes about Chaplin abruptly and
rapidly appear. Positive thoughts and feelings for Chaplin are
blocked, eliciting interest in the events which follow.
Individuals are drawn into the story, waiting for resolution --
either reinstatement of the positive attitude about Chaplin, or
until such scenes become so negative that the viewer's attitude
is altered.
Evidence of this conception of interest has been
reported. Boselie (1983) concluded that the presence of
disjunctive ambiguity in drawings were judged more interesting
because it induced conflicting views, compelling individuals to
search for new interpretations of an otherwise normal picture.
Berlyne and Boudewijns (1971) found that interest in a visual
pattern was related to the number of differences between
successive visual patterns. When small changes between successive
patterns occurred, processing tendencies were reinforced and
continued unhindered and these patterns were judged less
interesting. When differences between successive patterns were
marked, processing tendencies were arrested, increasing interest
in the pattern.
Event-related desynchronization
Temporary increases in attention or cognitive processing
due to response inhibitions may be identified in the continuous
EEG signal. One technique for assessing abrupt activation in
short segments of EEG is the event-related desynchronization
(ERD) paradigm. The ERD paradigm refers to quantification of
alpha amplitude for very brief intervals (125 ms) during repeated
sensory presentations and interstimulus intervals (Pfurtscheller,
1977). An ERD is a temporary suppression of the alpha rhythm,
lasting from 250 to 500 ms, which occurs which subjects process
information (Klimesch et al., 1992; Klimesch et
al., 1990). Immediately following an ERD, there is often
a momentary burst of alpha activity (Pfurtscheller, 1992; Veigel
& Sterman (1993). This period of deactivation resembles a
refractory response to preceding activation and is called post-
response synchronization, or PRS. No study has yet explicitly
investigated performance or processing correlates of PRS
magnitude; however, mean alpha activity in continuous EEG is
probably a function of ERD and PRS magnitudes. Averaging alpha
activity across a task, though, obscures this fact as each ERD-
PRS sequence is brief, lasting from 3 to 8 s (Pfurtscheller,
1992).
The concept of multiple ERD and PRS sequences during a
task is an apt description of arousal jags. Momentary arousal
(ERD) is followed by arousal-reduction (PRS), the rewarding
component of the jag (Berlyne, 1971). Sterman, Wyrwicka, & Roth
(1969) identified a similar post-reinforcement synchronization
after food or water consumption. Unfortunately, the ERD paradigm
requires repeated presentations of like stimuli, whereas everyday
cognition and most experimental tasks involve ongoing process.
Continuous EEG may consist of nothing more than a series of ERD
and PRS sequences, but the intervals between successive ERDs or
PRSs are unknown and uncontrolled by the experimenter for the
continuous EEG paradigm. One might attempt to identify ERD-PRS
sequences in continuous EEG by means of pattern recognition
algorithms, but it is presently unclear whether the duration of
ERD and PRS and the interval between are constant or vary as a
function of cognitive response, modality, and subject or stimulus
variables.
Notwithstanding these obstacles, ERD-PRS sequences can
be indirectly assessed by behavioral measures. Mean variability
in behavioral (joystick) interest ratings may also contain
relative information about ERD-PRS sequences. It is predicted
that variability of behavioral interest ratings gauge the
incidence of arousal "jags" and subsequently the degree of
subjective interest in a film.
Modulation of arousal
As mentioned in Chapter 5, arousal alone is not
responsible for eliciting interest. Mildly noxious stimuli may
produce arousal without interest (Zillman, 1982). Films which
consist of unending chases and gunplay that constantly attempt to
arouse the audience can become as monotonous as a blank screen.
Likewise, films with uneven pacing or tedious plotting can
diminish arousal and bore an audience. Interesting films required
arousing events as well as relatively uneventful intervals during
which individuals can integrate information (from previous
intense segments) as well as experience the pleasurable drop in
arousal.
Arousal levels oscillate not only during high interest
conditions, but also during low involvement conditions such as
eyes open and eyes closed baselines (cf. Figure 3.1). What
differs between the modulation pattern found in baseline
conditions (i.e., the natural background of alpha fluctuation)
and the modulation pattern during high interest states is the
source or origin of the modulation. When subjects relax with eyes
closed, so that they are no processing external stimulation, the
pattern of alpha modulation -- that is, the duration and
magnitude of specific ERD-PRS series -- will depend solely on the
subject's internal thought processes (reverie, ruminations,
etc.). When subjects pay attention to a stimulus in the
environment like a film, it is theorized that the information
pattern of a stimulus will drive the pattern of alpha modulation
and it should be possible to quantify interest in a film by
measuring the magnitude and duration of ERD-PRS sequences in
response to each cinematic event. Unfortunately, because
cinematic narratives do not consist of specific, short, non-
overlapping events (required for the ERD paradigm) to which all
subjects respond, another method for identifying and assessing
ERD-PRS sequences must be found.
Attentional synchrony
Individuals will exhibit ERDs to stimuli which interest
them. When few or no individuals display ERDS in response to a
scene in a film, it is consequently uninteresting. Boring scenes
will result in various responses from individuals as some
continue to process previous information and others anticipate
new events and or daydream, the important point being that they
are no responding in unity to the film. When individuals respond
in unity to a scene, demonstrating high attentional synchrony
(ASync), that scene must be interesting. When individuals are
interested in a narrative, they will increase attention as
suspense builds and relax during momentary breaks in the action
and so forth. Alpha activity will become parallel the pattern of
information presented in the film, increasing as it increases and
diminishing as it decreases. Unfortunately, as mentioned above,
assessment of individual attentional variation in reaction
to specific inputs is rather poor in a continuous task like film
presentations. But assessment of group synchrony,
simultaneous engagement or disengagement in each epoch, is
feasible.
One may acquire EEG from an individual at a time when he
or she is actively assimilating information (ERD) or after such a
process has occurred (PRS) or somewhere in between. It is
possible for 20 individuals to be at 20 different positions in
the ERD-PRS sequence at any given epoch. This situation occurs
when individuals in a group are involved in numerous, unrelated
tasks -- for instance, some individuals are solving math
problems, some are speaking, some are throwing a ball, and so
forth. If, however, all 20 subjects are performing the same task
at the same time and each subject still exhibits a different
position in the ERD-PRS sequence, it must be the case that some
individuals are attending to task elements while others are
engaged by earlier or later elements (but not the present one)
and still others are largely disengaged by the entire task.
Whether a group of subjects are synchronized in integrating a
narrative can be measured by between-subjects variance of alpha
magnitude. Small between-subject variance during each epoch
indicates processing homogeneity. Processing homogeneity can only
occur consistently if subjects are encoding the same
pattern of information at the same time. If some or all subjects
are uninterested and are more engaged by internal processes than
the shared external stimulation, the alpha magnitude exhibited by
each subject should be unrelated to the alpha magnitude in any
another and between-subject variance will be high.
When an individual is interested in a film, his or her
attentional state will change is result of narrative elements.
Essentially, one's attentional states (and associated alpha
activity) will be mediated by the external stimulation. If most
or all subjects are interested in a film, between-subject
variance of epoch magnitude will be small. As subjects grow
bored, between-subject variance will increase. The validity of
this measure can be partly verified by comparing it to subjective
interest values.
Low involvement conditions such as the eyes open
baseline displays high values of between-subject variance (see
Figure 6.1). Not only is the between subject alpha variance large
at every epoch, note the variability between subjects is
increasing as time progresses. At task onset, subjects were in a
similar psychological, behavioral, and motivational state, and
thus variance was relatively low. They were instructed to
close their eyes and relax, most acknowledged
verbally to these instructions, and they all sat back and
attempted to relax. But as time continued, their thoughts and
actions diverged as no external stimulation was forthcoming to
activate or mediate attentional states.
Figure 6.1. Mean alpha magnitude +/- 1 SD from 18
subjects are plotted for each epoch. Individual bars indicate the
range across subjects at each epoch during an eyes closed
condition (128 s, site Pz). Note the wide range of magnitudes at
any given epoch.
During low involvement conditions such as eyes open
baselines, alpha activity in each subject should be randomly
correlated to alpha activity in any other subject. When subjects
process the same stimulus to similar degrees, alpha activity in a
subject will show a positive correlation with alpha activity in
other subjects. Thus, higher correlations should indicate greater
group interest in a stimulus.
Behavioral synchrony is well established (e.g.,
interactional; Bernieri, Reznick, & Rosenthal, 1988; Maxwell,
Cook, & Burr, 1985). IPS refers to occasions when a group of
individual exhibit similar or converging alpha activity in
response to a stimulus. A reduction in between-subject variance
of alpha amplitude at specific cortical sites indicates high ASync
and conversely an elevation in between-subject variance indicates
low ASync. Large between-subject variance occurs when some subjects
experience an ERD at the same time that others undergo a PRS and
still others remain in a relatively neutral state. Relatively low
variability between subjects can only occur when alpha amplitude
in most or all subjects is increasing or decreasing at the same
time.
The functional areas wherein the greatest across-subject
synchronization occurs may depend on stimulus modality and
properties. It is predicted that higher ASync (reduced between-
subject variance) will be observed in parietal areas during high
interest narratives due to attentional aspects of story
integration (see Chapter 5). Reduced ASync is expected for
conditions such as eyes open baseline and low interest films.
Temporal dynamics of interest
ASync can be assessed across entire films, but it should
also be valid for short intervals of time. In this study,
subjects rated interest in a film by a subjective and by a
behavioral method. The behavioral technique (joystick rating)
provided a continuous record of interest as it varied across
time. Both alpha magnitude and ASync can be calculated for each
epoch and compared to the behavioral rating for the epoch. It is
predicted that both alpha magnitude and between-subject variance
(ASync) will be negatively correlated with behavioral interest
ratings. Again, it is expected that the best correlation will be
in parietal cortex.
Summary of predictions
A significant positive correlation is predicted between
variability of behavioral ratings and mean subjective ratings.
Mean ASync is predicted to differ between high and low interest
films. In addition, a significant correlation between mean ASync
and subjective ratings is predicted. Finally, epoch values of
alpha amplitude and ASync will correlate with behavioral ratings.
METHOD
The method described in Chapter 5 was used. Joystick
values were analyzed for 19 subjects. One subject failed to
follow instructions and was excluded from analysis.
RESULTS
The interval between joystick movements was 7.55 s on
average. As shown in Figure 6.2, a significant positive
correlation was found between mean subjective (Likert-type scale)
ratings and behavioral (joystick) ratings [r=.46,
F(1,18)=4.714, p<.05].
Figure 6.2. Correlation between joystick position
variability and subjective interest ratings (n=19, 20 films)
ASync and subjective interest
Examples of between-subject variance in low and high
interest films are depicted in Figure 6.3.
Figure 6.3. Mean alpha magnitude +/- 1 SD at Pz are
plotted for each epoch. Individual bars indicate the range of
alpha magnitude across subjects at each epoch during low and high
interest films. Note the narrow distribution of magnitude values
during the high interest film.
Mean between-subject SD was calculated for the entire
film (approximately 2 min) and for the last 30 s of each film in
11 lower interest and 10 higher interest films. Significant
topographic interactions were found [FLASyncT30s(6,110)= 3.440,
p<.01; FENTIRE(6,111)= 2.968, p<.01]. As shown
in Figure 6.4, mean between-subject variance was lower for high
interest films compared to low interest films at sites Cz and Pz
for the last 30 s of a film [p<.01].
Figure 6.4. Mean between-subject variance for 10 higher
and 11 lower interest films. Variance was measured across entire
films or for the last 30 s (* p<.01).
As shown in Figure 6.5, a significant negative
correlation was found between normalized subjective interest
ratings and between-subject variance during the last 30 s of each
film at site Pz [r=-.80, FLASyncT30s(1,19)= 33.627,
p<.01]. A significant relationship was also found at site
Cz [r=-.77, FLASyncT30s(1,19)= 27.831, p<.01].
Figure 6.5. Mean between-subject variance as a function of
(normalized) subjective interest ratings. Numbers in graph refer
to EO conditions placed arbitrarily as low interest conditions.
Comparison of ASync and magnitude values
Behavioral rating, ASync, and alpha magnitude, were
calculated at site Pz for each epoch of each film. Behavioral
ratings were compared to between-subject variance at each epoch
(ASync) and epoch magnitude for eight films. Only eight films had
reliable behavioral ratings (i.e., ratings from six of more
subjects). A 50 s smoothing function and a 6 s time lag of ASync
provided the best correlation with behavioral ratings. Of the
eight films analyzed, six exhibited significant correlations
between behavioral ratings and ASync values. Only 4 films exhibited
a significant correlation between behavioral ratings with epoch
magnitude [p<.05]. Figure 6.6 illustrates the relationship
between magnitude, between-subject variance, and behavioral
ratings across time.
Figure 6.6. Epoch values of between-subject variance,
alpha magnitude, and a behavioral interest rating during two
films [Cool World, Heart of Dixie]. Note the especially strong
relationship between either EEG measure and the interest rating
in the second film.
DISCUSSION
Behavioral and subjective interest ratings correlated
with each other. A relationship between subjective interest and
mean between-subject variance was found at specific sites. A
correlation between behavioral interest ratings and between-
subject variance was found across time in the majority of films
examined.
Behavioral index of arousal jags
As arousal increases in response to events in a film,
the subject performing the continuous rating task was expected to
move the joystick as interest increased. When arousal
subsequently declines, returning to a previous level, the subject
was expected to change the joystick position a second time. The
number of arousal jags should thus be reflected in the
variability of joystick movements performed by each subject.
Because interesting stimuli generate more arousal jags
than boring stimuli, high interest films were predicted to result
in higher variability in the behavioral rating task. The
relationship was modest (r=.46), however. This may be due to the
very limited range of movements allowed by the joystick. Subjects
were instructed to assess interest continually and may have felt
that any change in interest required a joystick movement. Given
the limited range of positions, subjects appeared to have used
most of the range for all films, inflating the rating variance
for low interest films.
Mean attentional synchrony
Mean ASync at mid-central and mid-parietal cortex varied
as a function of subjective interest. Mean ASync for the last 30 s
of a film differentiated higher and lower interest films. Mean
ASync for the entire film also distinguished these films but to a
lesser extent (p=.02). ASync is less predictive across the
entire film as individuals pay attention at the film onset,
regardless of interest (Hawkins et al., 1991).
Electrical activity in central cortex is often used to
assess physical relaxation (e.g., Borgeat, 1983). Arousal during
suspense films is associated with changes in bodily sensations
(Hubert, de Jong, & Meyer, 1991). ASync values at mid-central
cortex may indicate overall muscular tone and tension associated
with arousal. Stratton and Zalanowski (1984) reported reduced
relaxation during atonal music. High interest films also resulted
in reduced between-subject variance (high ASync) at mid-parietal
cortex, but this effect is attributed to cognitive components of
subjective interest.
Iran-Nejad and Cecil (1992) claimed that story interest
is related to post-surprise intellectual activity. High-surprise
story endings are more interesting than low-surprise story
endings as long as post-surprise incongruity is resolved. While
identification of surprising elements is a left hemisphere
function, integrating this unexpected element into an ongoing
narrative may be performed by the right hemisphere. Bihrle,
Brownell, and Powelson (1986) tested comprehension of humorous
materials by left and right brain-damaged patients. Right
hemisphere patients show a preserved sensitivity to the surprise
element inherent in humorous material but a reduced ability to
establish narrative coherence from the material. Conversely, left
hemisphere patients had trouble recognizing the surprise element
of a humorous sequence, but were able to establish coherence by
integrating multiple elements into a single narrative. Brownell
et al. (1983) also reported left and right hemispheric
specialization for surprise and coherence elements of humor
processing, respectively.
Very interesting stimuli engage both hemispheres, as
shown in Chapter 5, but the pattern of activation may differ
between individuals. A moment of suspense might engage analytic
or verbal processes in one subject while another subject is
engaged in visuospatial or holistic functions. Although the left
and right parietal cortex activate during very interesting scenes
in all subjects, the actual timing of left and right hemisphere
functions may differ between individuals and therefore ASync is not
as large in these areas as it could be (i.e., if all subjects
demonstrated similar cognitive styles). Alpha magnitude at Pz,
however, reflects activity in both left and right parietal lobe;
regardless of which hemisphere is activated at any given moment,
Pz will be activated, if somewhat indirectly.
An argument might be made that ASync findings are actually
a "floor effect" in alpha magnitude: lower amplitudes generally
have diminished variability and lower amplitudes are associated
with higher interest. However, the relationship between alpha
magnitude and subjective interest ratings (r=-.42) is not as
strong as that found for ASync (r=-.65). Also, ASync is often
inversely related to magnitude values in some films (see Figure
6.4).
The reliability and validity of this new measure, ASync,
has not yet been established. However, the present results are
very promising that this approach will yield useful information
about conditions that other EEG measures cannot obtain. Mean ASync
and epoch ASync values were both highly predictive of subjective
interest in films. The one drawback to this measure is that it
cannot be measured for an individual but requires a group.
|
