Perceptual learning on inspection time 1

Running head: INDIVIDUAL DIFFERENCES ON PERCEPTUAL LEARNING

Perceptual learning on inspection time and motion perception

Nicholas R. Burns, Ted Nettelbeck and Jason McPherson

University of Adelaide, Australia

Lazar Stankov

University of Sydney, Australia.

Perceptual learning on inspection time 2

Abstract

Perceptual learning on simple perceptual tasks is interpreted as plasticity of neuronal

populations in sensory cortex (Fahle, M. & Poggio, T., 2002). This paper examines

individual differences on perceptual learning for two tasks, inspection time (IT), and a

motion direction discrimination task instantiated as random dot kinematograms. The main

questions were whether individual differences in perceptual learning were consistent across

the two tasks; and whether perceptual learning correlated with cognitive abilities test scores.

N = 56 young adults completed 16 threshold estimations on one of two orthogonal versions of

each task. Two further threshold estimations were then made for the untrained, orthogonal

version. Participants also completed a battery of six cognitive abilities tests measuring fluid

ability (Gf) and perceptual speed (Gs). Perceptual learning was demonstrated for both tasks

but the degree of learning across tasks was not characteristic of the individual. Learning on

IT correlated with Gs (r = .35) but learning on the motion direction discrimination task was

unrelated to cognitive ability. Correlations of IT with cognitive measures were stable over the

training period. IT was correlated with both the motion direction discrimination task (r = -

.39) and with an unmasked line length judgment task (r = -.31). We concluded that perceptual

learning on IT correlates with cognitive abilities test scores, that correlations of IT with

cognitive abilities test scores are stable as task performance improves with practice, and that

the IT task is psychologically complex.

Keywords: Perceptual learning; Inspection time; Motion perception; Cognitive ability

Perceptual learning on inspection time 3

Perceptual learning on inspection time and motion perception

In the search for biological correlates of individual differences in human intelligence,

the net has been cast widely. Thus, Vernon’s (1993) edited volume contained chapters on, for

example, cerebral glucose metabolism, biochemical correlates, EEG and evoked potential

studies, and physical correlates. However, a recurrent theme in Deary’s (2000) book on

intelligence was that establishing such correlations needs ultimately to be accompanied by

well-formulated explanations for mechanisms causing correlations. At present it is difficult to

link psychological level explanations with biological level accounts. This is particularly the

case when considering individual differences in complex behaviors required to solve items

from an intelligence test. Nonetheless, since Spearman’s (1927) suggestion of mental energy

as the basis for general ability (i.e., g), many such explanatory proposals have been made.

Recently, a general property of the brain suggested as a candidate for explaining superior

cognitive performance is neural plasticity (e.g., Garlick, 2002; Simonton, 2003).

It is clear that the brain exhibits remarkable plasticity over multiple different scales

and time frames. The reorganization of somatosensory cortex seen in amputees and studied

extensively in animal models is an astonishing example (Ramachandran & Hirstein, 1998).

Less spectacular but more accessible to study is the phenomenon known to psychophysicists

as perceptual learning. Perceptual learning refers to stimulus specific improvement in the

ability to perform simple perceptual tasks (Ball & Sekuler, 1987; Fiorentini & Berardi, 1980;

Karni & Sagi, 1993; Poggio, Fahle, & Edelman, 1992; Ramachandran & Braddick, 1973).

This has been interpreted as demonstrating plasticity at early stages in the cortical

information processing hierarchy because, as Zanker (1999, p.1293) noted, “improvement is

not just a consequence of getting acquainted with an experimental setting in a general sense,

nor does it result from a slow understanding of the task.” Indeed, the specificity of perceptual

learning on visual tasks to certain stimulus parameters (e.g., direction of motion, motion type,

Perceptual learning on inspection time 4

orientation, size) suggests specific mechanisms underlying the demonstrated cortical

plasticity. For example, Zohary, Celebrini, Britten and Newsome (1994) recorded from single

cells in area MT of monkeys over the course of an experiment in which the monkeys were

trained visually to discriminate direction of motion in random dot kinematograms (RDKs).

They found improved performance on the task (i.e., a decrease in motion coherence

thresholds) over successive blocks of trials, accompanied by a corresponding increase in

directional sensitivity for single neurons in MT. They concluded that, in principle, perceptual

learning could arise as a result of increased sensitivity of sensory neurons.

It once seemed surprising that plasticity could be observed at such early stages in

visual processing because necessary connections were thought to be hard-wired. However,

extracting, say, motion from noisy signals is a critical task for everyday vision; for example,

driving a motor vehicle through fog or rain. Thus, some functional plasticity of early visual

processing mechanisms clearly has considerable adaptive value. It also once seemed

surprising that individual differences in performance on simple perceptual tasks would

correlate with scores on abilities tests. Interestingly, mechanisms proposed to explain both

these phenomena have at various times invoked similar processes. Computational models of

perceptual learning involve noise parameters and the concept of suppression of both noisy

inputs and the internal noise inherent to biological systems (Dosher & Lu, 1999). Certain

biological models of intelligence have described noise-free transmission of information

within the nervous system as determining superior mental performance (Eysenck, 1982).

Inspection time (IT) is a perceptual task with low knowledge requirements that

correlates with intelligence (Nettelbeck, 1987). Two high-contrast lines are presented side by

side, followed after a variable duration by a pattern mask that overlies the target stimulus.

Thus, IT is a visual backward pattern masking task, the task requirement being to nominate

which target line was shorter. Some studies have reported that IT performance among

Perceptual learning on inspection time 5

children improves with practice (Nettelbeck & Wilson, 1985; Nettelbeck & Vita, 1992).

Among adults, Chaiken (1993) found improvement on IT over two estimations separated by

about two hours but mainly at short rather than longer SOAs. Sadler and Deary (1996)

reported strong learning effects on IT over five days of testing and the effect was present at

all SOAs across fully characterized psychometric functions. Stokes and Bors (2001) also

reported strong learning effects over five days on an IT task that used letters rather than lines

as the target stimuli. Nettelbeck, Rabbitt, Wilson and Batt (1996) reported a statistically

significant practice effect on IT among elderly participants across 18-20 months. Perceptual

learning has been demonstrated for some visual masking tasks (Dorais & Sagi, 1997) but

there are no reports of perceptual learning on IT.

Interestingly, much debate on the nature of IT and its correlation with intelligence has

centered on what perceptual processes determine performance on the task. The original task

was framed in terms of the comparison of the length of two lines (Vickers, Nettelbeck, &

Willson, 1972). More recently, it has been speculated that performance depends on the

detection of motion cues generated when the target figure is replaced by the mask figure

(Burns, Nettelbeck, & White, 1998; White, 1996). Perceptual learning for the detection of

motion and discrimination of motion direction has been demonstrated by Ball and Sekuler

(1987) using sparse dot RDKs and by Zanker (1999) using dense dot RDKs. The primary

finding was that perceptual learning was for direction of motion. Thus, learning transferred to

directions close to the trained orientation but not to other directions. Some research has

examined the detection and discrimination of motion direction in relation to individuals with

very low levels of intelligence compared to normal controls (Carlin, Hobbs, Bud, & Soraci,

1999). These researchers reported improvement on motion direction discrimination with

practice on the task even for participants with intellectual disabilities. However, such

individuals required detailed instruction and feedback during training. Investigation of

Perceptual learning on inspection time 6

perceptual learning on IT thus may shed light on the mechanisms underlying performance on

the task and on its relationship with intelligence.

This paper reports an experiment in which participants were trained on either one of

two versions of IT (the traditional version with two vertical lines, or an orthogonal version

consisting of two horizontal lines). For both versions the task was to nominate which line was

shorter; left or right for the former, top or bottom for the latter. Participants were also trained

on either one of two versions of a motion direction discrimination task (motion was in a

vertical plane or a horizontal plane). For both versions the task was to nominate the direction

of motion; up or down for the former, left or right for the latter. After a training phase,

participants were tested on the alternative untrained, orthogonal version of each task.

Evidence for perceptual learning is an improvement of performance during the training

phase, accompanied by a decrement in performance on the untrained, orthogonal version of

each task, relative to the level attained during training. A question we address in the current

paper, then, is whether there is perceptual learning on IT. Consistent with previous research,

we expected to demonstrate perceptual learning on the motion direction discrimination task.

Of interest was the extent of individual differences on perceptual learning for both tasks and

whether it was characteristic of the individual. We asked whether perceptual learning on

either task correlates with intelligence test scores, the expectation being that magnitude of

improvement would correlate positively with test scores. The inference would be that degree

of neural plasticity was related to superior mental performance. We also examined the pattern

of correlations of IT with intelligence test scores during the course of training; the literature

is equivocal on this issue, with some reports that correlations increase in magnitude as

experience with IT increases but other reports of just the opposite. Finally, we explored IT in

terms of putative components, specifically, line length judgment and the detection of motion

signals generated by the backward masking procedure used in the IT task. Does superior

Perceptual learning on inspection time 7

performance on an unmasked line length judgment task, or on the motion direction

discrimination task, or on both of them, determine IT?

Method

Participants

There were N = 56 participants (23 males; mean age = 24.2, SD = 6.6 years) recruited

via an advertisement placed with the University employment service. Most were students and

all were paid A$50 at completion of the study.

Apparatus and materials

For both the inspection time and motion direction discrimination tasks stimuli were

displayed at a viewing distance of 1 m on a high-resolution monitor (1024 x 768 pixels, 100

Hz vertical refresh rate) driven by a digital stimulus generator (VSG 2/3; Cambridge

Research Systems).

Inspection time. For the traditional IT task, hereafter referred to as ‘Vertical IT’, the

target figure consisted of two vertical lines, one subtending a visual angle of 2.1 deg and the

other 2.5 deg and joined at the top by a horizontal line subtending an angle of 1.2 deg. The

shorter line appeared on the left or right equiprobably. Each trial involved the following

sequence of events. The target figure was preceded by a warning beep. Following exposure

of the target figure for the relevant stimulus onset asynchrony (SOA, see below), it was

immediately replaced by a “flash” mask (Evans & Nettelbeck, 1993) of 300 ms duration and

consisting of two vertical lines subtending a visual angle of 3.3 deg and shaped as lightning

bolts (see Figure 1). For the orthogonal version of the IT task, hereafter referred to as

‘Horizontal IT’, details were the same as for Vertical IT but the target figure consisted of two

horizontal lines joined at the left by a vertical line; the shorter line appeared on the top or

bottom equiprobably. The participant indicated the location of the shorter line in the target

Perceptual learning on inspection time 8

figure by pressing the appropriate button on the computer mouse, the orientation of which

was changed to suit the Vertical IT and Horizontal IT tasks.

Motion direction discrimination. There were two versions of this task, hereafter

referred to as ‘Vertical Motion’ and ‘Horizontal Motion’, respectively, and instantiated as

dense dot RDKs (see Figure 1). For both versions the stimulus field was 256 pixels x 256

pixels (10.8 x 10.3 deg). Within a dynamic noise background consisting of black and white

dots, a 16 x 16 pixels square group of black and white dots was either displaced up or down,

for Vertical Motion; or displaced left or right, for Horizontal Motion (direction of motion was

equiprobable). This motion-defined object was shifted by one pixel between frames (motion

at about 3 deg/sec) and its starting position was randomized within an area of 32 x 32 pixels

centered on the middle of the stimulus field. On any trial, a proportion of dots within the

motion-defined square was inverted between frames (black pixels replaced by white and vice

versa; inverted dots were randomly chosen with the proportion determined by the estimation

algorithm, see below). This manipulation allowed the signal-to-noise ratio of the motion-

defined object to be systematically varied in steps of 10% noise from 0-to-100% noise. When

the proportion of dots exchanged becomes high the object is no longer perceived against the

dynamic noise background (where the probability of any dot being black or white on any

frame is 0.5). Each trial commenced with the appearance of 12 frames of the dynamic noise

background, followed, after a brief beep, by the 12 frame stimulus sequence and ended with

another set of 12 noise frames. For the Vertical Motion task, participants indicated whether

the square moved up or down and for the Horizontal Motion task, whether it moved left or

right, by pressing the appropriate button on the computer mouse, which was oriented so that

the keys aligned with left-right or up-down configurations.

Psychometric tests. Fluid ability (Gf) was assessed using two tests from the

Woodcock-Johnson Psycho-Educational Battery-Revised (WJ-R; Woodcock & Johnson,

Perceptual learning on inspection time 9

1989), Analysis-Synthesis, which involves solving for the missing components of an

incomplete logic problem, and Concept Formation, a test of categorical reasoning. General

processing speed (Gs) was assessed using Visual Matching and Cross Out, also from the WJ-

R and both measuring clerical speed. Broad visualization ability (Gv) was assessed using two

computerised tests from Stankov’s Tests of Cognitive Ability (STOCA), a battery used for

the selection of train drivers in New South Wales (Australia). In the Line Length test, five

vertical lines are presented on the screen in a staggered fashion. The task is to indicate which

line is longer than the remaining four. The length of the odd line differs from the rest by from

one to five pixels. In the Letter Spotting test, a set of ten letters is presented on the screen for

800 msec, followed by another set of ten letters, and so on, until ten sets of letters have been

presented. Some of these letter sets contain a designated letter (e.g., the letter “o”). The task

is to count the number of times this letter appeared on the screen. For tests from WJ-R, W-

scores (Rasch-scaled scores, centered on 500) were used; for tests from STOCA, scores were

percentage items correct.

Procedure. Participants attended five sessions in the laboratory spread over a period

of at least one week but no more than two weeks. The first session lasted about 90 min and

subsequent sessions lasted about 60 min. At the first session, participants were assigned to

one of the four possible combinations of the two psychophysical tasks. Task requirements

were explained using unmasked stimuli for IT and zero noise stimuli for motion direction

discrimination. Once participants understood task requirements, including the instruction that

accuracy not speed of response was important, threshold estimations commenced. To

determine psychophysical threshold we used an adaptive stair case algorithm (Wetherill &

Levitt, 1965) that returned an estimate of the 79% correct performance level. Thus, we used a

three-up, one-down rule for determining the SOA (or noise level) presented on any trial. That

is, three correct responses were required before reducing the SOA by 10 ms (or increasing the

Perceptual learning on inspection time 10

noise level by 10%) but SOA was increased by 10 ms (or noise level decreased by 10%) after

each incorrect response. For each estimation, the initial trial had an SOA of 250 ms (or noise

level 0%). Estimated threshold was the average of 12 reversals on the staircase. Two

estimations for each task were completed at this first session. Following this, all

psychometric testing was completed. At the next three sessions, four estimations for each

psychophysical task were completed (estimations for each task were completed alternately).

At the final session, two estimations for each task were made. Then, two further estimations

were completed but on the orthogonal orientation to that for which previous estimations had

been made. Thus, a participant who had completed 16 estimations for Vertical IT and

Horizontal Motion now completed two estimations each on Horizontal IT and Vertical

Motion.

Results

Figure 2 shows mean threshold performance for each task at each of the 18 separate

staircase estimates. The metric for the dependent variable was different for the IT tasks and

the motion direction discrimination tasks. For the former, SOA (in msec) was the dependent

variable and for the latter it was the proportion of noise (percentage noise) added to the

square group of dots moving against the dynamic noise background. Thus, the two tasks were

analyzed separately using repeated measures analysis of variance where the successive

threshold estimations during the training phase of the experiment were the 16 levels of the

independent variable. There was also a between-subjects factor with two levels; that is, the

version of the task used during the training phase of the experiment (Vertical or Horizontal).

We note here that we adopted the multivariate approach to repeated measures analysis of

variance. This approach treats the set of repeated measures together and has a major

advantage over the traditional approach in that it makes no assumptions on the sphericity of

the variance-covariance matrix. This multivariate approach is marginally less powerful than

Perceptual learning on inspection time 11

the traditional approach but this disadvantage is outweighed because fewer assumptions are

made (see e.g., Howell, 2002).

For IT, the ANOVA showed that there was no effect of training group membership,

F(1, 54) = 2.29, p = .14. The training group by staircase estimate interaction was also not

statistically significant, F(15, 40) = 1.44, p = .18, but the effect of training was, F(15, 40) =

8.84, p < .001, partial η2 = .77. Thus, performance on both Vertical and Horizontal versions

of IT improved over the 16 estimates of the training phase (i.e., IT estimates became shorter).

This improvement encompassed statistically significant linear, F(1, 54) = 62.6, p < .001,

partial η2 = .54, and quadratic, F(1, 54) = 8.87, p = .004, partial η2 = .14, components.

To establish whether perceptual learning can be claimed on IT, it remains to be

demonstrated that there was a decrement in performance when participants were tested on the

orientation orthogonal to that for the training phase. The thresholds for the first and last two

staircase estimates in the training phase (hereafter Initial and Final) and the two estimates

from the test phase (hereafter Test) were averaged. These mean estimates are shown in Table

1. A repeated measures ANOVA, with time of testing now having three levels and with the

between-subjects factor as before, was conducted. The ANOVA showed that there was no

effect of training group membership, F(1, 54) = 0.14, p = .71. However, the training group by

time of testing interaction was statistically significant, F(2, 53) = 13.1, p < .001, partial η2 =

.33, as was the effect of time of testing, F(2, 53) = 39.9, p < .001, partial η2 = .60. The

interaction arises because of between group performance differences at Initial and Test (see

Figure 2 and Table 1). The effect for time of testing, along with the means shown in Table 1,

establishes perceptual learning for IT.

For motion direction discrimination, the ANOVA, conducted as for IT, showed that

there was no effect of training group membership, F(1, 54) = 0.19, p = .66. The training

group by staircase estimate interaction was also not statistically significant, F(15, 40) = 0.96,

Perceptual learning on inspection time 12

p = .52 but the effect of training was, F(15, 40) = 7.96, p < .001, partial η2 = .75. Thus,

performance on both Vertical and Horizontal versions of IT improved over the 16 estimates

of the training phase (i.e., noise level estimates improved). This improvement encompassed

statistically significant linear, F(1, 54) = 85.3, p < .001, partial η2 = .61, and quadratic, F(1,

54) = 14.2, p < .001, partial η2 = .21, components.

To establish whether perceptual learning can be claimed on motion direction

discrimination, the same procedure was followed as for IT. Mean estimates for Initial, Final

and Test are shown in Table 1. The ANOVA showed that there was no effect of training

group membership, F(1, 54) = 0.08, p = .78, and the training group by time of testing

interaction was also not statistically significant, F(2, 53) = 2.02, p = .14. The effect of time of

testing was statistically significant, F(2, 53) = 32.6, p < .001, partial η2 = .55. The effect for

time of testing, along with the means shown in Table 1, establishes perceptual learning for

IT.

Examination of individual data showed clear differences across participants in their

patterns of learning for both tasks. However, quantifying the extent of this learning was

difficult because of variability within individuals across both estimations within sessions and

across sessions. In an attempt to capture the extent of learning across the training phase, data

were smoothed via a three-point moving average procedure. Improvement was then

quantified as the change in performance between the first datum point and the last datum

point expressed as a percentage of the first datum point. Individuals trained on Horizontal IT

improved more than those trained on Vertical IT (29.9% (SD = 31.6) and 20.0% (SD = 18.0,

respectively) but this difference was not statistically significant (Welch’s t(adjusted df = 42)

= 1.44, p = .16, d = 0.42). Individuals trained on Horizontal Motion improved more than

those trained on Vertical Motion (19.3% (SD = 17.0) and 12.9% (SD = 15.7) but again this

difference was not statistically significant (Welch’s t(adjusted df = 53) = 1.46, p = .15, d =

Perceptual learning on inspection time 13

.39). Correlations between learning on the two tasks, whether considered for the sample as a

whole, or for subsets defined by training direction, were near zero and not statistically

significant. Thus, there is no learning parameter for each individual that applied across the IT

and Motion tasks. This outcome was replicated when various functions were fitted to

individual learning curves for both tasks.

Turning now to correlations of perceptual learning with different cognitive ability test

scores. First, a principal components analysis of the six abilities measures was performed.

The first component accounted for 34% of variance and the second component accounted for

a further 26% of variance; these components were the only two with eigen values greater

than unity. An oblique rotation (promax) of these two components was easily interpretable.

One component was Gf, loaded by Analysis-Synthesis and Concept Formation. The other

was loaded by the remaining four tests and is clearly Gs. It seems that in this sample

visualization ability was too homogeneous to enable definition of a separate Gv component.

Table 2 shows the descriptive statistics for the cognitive abilities data, along with the

correlation of each test, and of the component scores for Gf and Gs, with the improvement on

each task. These data are for the sample as a whole and we noted that there were no

statistically significant differences on cognitive ability test scores for any of the tests across

the four training groups. The only correlations of note were between learning on the IT task

and Cross Out from WJ-R and Letter Spotting from STOCA. These indicate a positive

correlation between improvement on the IT task and performance on tests of Gs.

Table 3 shows the correlations between Vertical IT and Horizontal IT and Gs and Gf,

as defined above, at each staircase estimate during the training phase. For Vertical IT,

correlations were consistent with expectations. That is, correlations were all negative and of

greatest magnitude for Gs. Unexpectedly, however, the correlations for Horizontal IT were

not statistically significant. For Vertical IT, the correlations with Gs do not appear to change

Perceptual learning on inspection time 14

reliably over the training period. The correlation of Vertical IT with Gf is largest at the first

three staircase estimates and then becomes close to zero as the training phase progresses.

There are no discernible patterns in the correlations for Horizontal IT.

The final issue addressed concerns whether performance on IT is accounted for by

performance on the motion direction discrimination task, or the unmasked line length

judgment task from STOCA, or both of them. The data considered are for the sample as a

whole and for the average of the first two staircase estimates for IT and the motion

discrimination task (IT Initial and Motion Initial, see above). The correlation between IT and

Motion was r = -.39, p = .003, CI95 = [-.59, -.14] and the correlation between IT and Line

Length was r = -.31, p = .021, CI95 = [-.53, -.05]. The partial correlation between IT and

Motion, controlling for Line Length, was -.36, p = .007; the partial correlation between IT

and Line Length, controlling for Motion, was -.26, p = .053. Thus, both motion direction

discrimination performance and accuracy of unmasked line length judgment make

independent contributions to variance in IT. This pattern of correlations held when Vertical

IT and Horizontal IT were analyzed separately, and if other estimates than the initial two

staircases were considered. It was also of interest whether partialling Motion and Line Length

from IT affected correlations of IT with ability test scores. We assessed this only for Vertical

IT because Horizontal IT did not correlate with Gs or Gf. In short, the zero-order correlations

of Vertical IT (average of first two threshold estimates) with Gs and Gf were -.32 and -.34,

respectively. Residualized Vertical IT correlated with Gs and Gf, -.25 and -.38, respectively.

That is, the correlations scarcely changed suggesting the correlation of IT with ability test

scores does not depend on either line length discrimination or motion discrimination.

Discussion

We have demonstrated perceptual learning on both IT and the motion direction

discrimination tasks. The latter finding confirmed previous research (Ball & Sekuler, 1987;

Perceptual learning on inspection time 15

Zanker, 1999) but the former is novel. The evidence for perceptual learning on IT was

stronger for the traditional version of the task, that is, the target with two vertical lines.

Participants initially found a different version of the task where the target was two horizontal

lines more difficult but, by the end of the training phase, performances on both versions were

identical. Then, a greater degree of transfer of learning to the untrained orientation was

observed for those in the Horizontal IT training group. This outcome meant that the

Horizontal IT group did not show as large a decrement in performance when tested on the

Vertical IT task as did the Vertical IT group when they were tested on Horizontal IT.

Previous research on IT which reported improvement with practice did not consider the

possibility that this improvement represented perceptual learning. That is, an implication of

our finding is that improvement in performance on the IT task arises due to changes in

perceptual mechanisms very early in the visual processing hierarchy. We discuss this point

further below when considering changes in correlations between IT and cognitive ability test

scores as IT performance becomes practiced.

We observed substantial individual differences in degree of improvement in

performance on both IT and the motion direction discrimination tasks. However, there was no

intraindividual consistency in degree of improvement across the two tasks. This null finding

should be viewed in the context of the fact that we have used only two visual tasks to study

perceptual learning from an individual differences perspective. Moreover, in this study we

sampled mainly university students, a group restricted in range on all cognitive abilities

dimensions. We therefore suggest that our study could be extended to a more representative

sample and that a range of simple perceptual tasks, including sensory modalities other than

visual, be used in an effort to quantify perceptual learning as a parameter that may represent

more general cortical plasticity. Such a parameter may at least partially define a general

intelligence factor.

Perceptual learning on inspection time 16

In the current study, improvement on the two tasks was difficult to quantify in a

refined way, with both tasks showing variability in performance both within sessions and

across sessions. This was particularly so for IT, with variability in mean performance for this

task, expressed as a proportion of mean performance, being much higher than for the motion

direction discrimination task. This suggests the possibility that IT is a more complex

perceptual task than the motion perception task, perhaps offering more variety of possibilities

to resolve the task, both perceptual and strategic. Certainly, IT was estimated reliably here

(median for one-back lagged correlations, that is, test-retest reliability over 16 estimations

were .79 and .80, for Vertical and Horizontal IT, respectively). Contrary to predictions based

on a strong version of the plasticity hypothesis, correlations of learning with cognitive

abilities test scores were generally low. However, for learning on IT, we did establish a

correlation with Gs of r = .35; and given the crudity of the learning measure adopted here,

this may be enough to encourage an attempt to refine the procedures we used. The lack of

correlation with learning on the motion direction discrimination tasks is plausibly due to

restricted variance on these tasks and their lack of reliability (median for one-back lagged

correlations, that is, test-retest reliability over 16 estimations were .50 and .57, for Vertical

and Horizontal Motion, respectively).

Previous evidence on the question of whether correlations of IT with intelligence test

scores change as performance on IT becomes more practiced is contradictory. Some

researchers have reported that practice on the task reduces correlations with intelligence test

scores (e.g., Nettelbeck & Vita, 1992). Stokes and Bors (2001) reported that with their

version of IT (actually a pattern backward masking task with two letters as targets and a

same-different discrimination), performance improved over five sessions where IT was tested

once each session. They also reported that the correlation between IT and a short version of

the Raven APM was .41 on Day 1 but decreased to .15, .18, .25, and .11, on the successive

Perceptual learning on inspection time 17

four days. However, the small sample size (N = 39) and consequent wide confidence intervals

for these correlations make it difficult to discern whether the correlations decreased reliably

with practice. Similarly, Bors, Stokes, Forrin and Hodder (1999) using a two lines IT task,

reported that performance improved over three sessions where IT was tested once each

session. Correlations between IT and Raven APM were -.43, -.15, and -.07, over the three

sessions, respectively. However, on an N = 17, these correlations are not statistically different

from each other. Chaiken (1993), on the other hand, reported that correlations of IT with the

Cattell Culture Fair Test increased from .35 to .50 across two estimations made about two

hours apart. Even with a much larger sample (N = 163), 95% confidence intervals around

these correlations overlap to such an extent that whether the correlation increased reliably is

doubtful. On balance, the conclusion we draw from these studies is that the correlation of IT

with intelligence test score may be unaffected by practice. Moreover, these studies used

measures of Gf or an IQ proxy but recent research points to IT measuring Gs and being

independent of Gf (Burns & Nettelbeck, 2003; Burns, Nettelbeck, & Cooper, 1999;

Crawford, Deary, Allan, & Gustafsson, 1998; Mackintosh & Bennett, 2002). For the current

study, Vertical IT showed correlations with Gs (mean r = -.39) that were remarkably stable

over 16 staircase estimates. Again, although not statistically different from each other, the

correlations with Gf (mean r = -.22) were highest for the initial staircase estimates. This latter

outcome, at least, is consistent with Stokes and Bors and Bors et al. We can offer no

explanation for the finding that Horizontal IT had near zero correlations with Gs (mean r = -

.02) and Gf were near zero (mean r = .03).

The final issue we addressed in this study concerned the nature of perceptual

processes involved in resolving the IT task. It was noted that performance on both motion

direction discrimination and unmasked line length judgment contributed to performance on

IT. However, for Vertical IT, partialling these measures from IT did not substantially affect

Perceptual learning on inspection time 18

correlations with Gs or Gf. Thus, at least two perceptual processes, line length judgment and

motion discrimination, contribute to IT performance; however, neither separately, nor

together, do they account for total covariance between IT and Gs or Gf.

To summarize, perceptual learning, which is taken to represent a form of plasticity at

early stages of the cortical processing hierarchy (see Fahle & Poggio, 2002), was observed

for both IT and motion direction discrimination. There were substantial individual differences

in the extent of perceptual learning but this was not characteristic of individuals across the

two tasks. Perceptual learning on IT correlated with Gs but perceptual learning on motion

direction discrimination did not correlate with cognitive ability test scores. Because we

demonstrated that improvement on IT arises due to perceptual learning, we can infer that

improvement on the task arises, at least in part, due to changes in neuronal populations very

early in the visual processing stream. This inference, coupled with the observation that the

correlation of IT with intelligence test scores is stable even though performance on IT

improves, has at least two possible interpretations. First, the locus of the correlation of IT

with IQ is later in the cortical processing hierarchy than the early visual processing stages

where line orientation perception and motion perception are localized. Second, the degree of

change in early visual stages responsible for learning on IT may be highly correlated with

whatever processes underlie the correlation of IT with intelligence. This second possibility

receives some support from the observation that learning on IT correlates with intelligence

test scores. Finally, the perceptual processes involved in IT are complex, involving both line

length judgment and the detection of motion. Many researchers have invested time in

attempting to reduce the strength of apparent movement cues available in IT when the mask

replaces the target (Burns et al., 1998; Chaiken & Young, 1993; Evans & Nettelbeck, 1993;

Knibb, 1992; Stough, Bates, Mangan, & Colrain, 2001). This may well be impossible and, in

any case, the detection of motion in the environment is an ability with such adaptive value

Perceptual learning on inspection time 19

that it is something that humans do, whether they report the use of such a strategy to resolve

the task, or not. Motion detection is likely an intrinsic part of IT and other pattern backward

masking tasks but might not be central to the correlation between IT and IQ.

Perceptual learning on inspection time 20

References

Ball, K., & Sekuler, R. (1987). Direction-specific improvement in motion discrimination.

Vision Research, 27, 953-965.

Bors, D. A., Stokes, T. L., Forrin, B., & Hodder, S. L. (1999). Inspection time and

intelligence: Practice, strategies, and attention. Intelligence, 27, 111-129.

Burns, N. R., & Nettelbeck, T. (2003). Inspection time in the structure of cognitive abilities:

Where does IT fit? Intelligence, 31, 237-255.

Burns, N. R., Nettelbeck, T., & Cooper, C. J. (1999). Inspection time correlates with general

speed of processing but not with fluid ability. Intelligence, 27, 37-44.

Burns, N. R., Nettelbeck, T., & White, M. (1998). Testing the interpretation of inspection

time as a measure of speed of sensory processing. Personality and Individual

Differences, 24, 25-39.

Carlin, M. T., Hobbs, K. L., Bud, M. J., & Soraci, S. A. (1999). Detection of motion-defined

forms by individuals with mental retardation and autism: Evidence of modifiability.

Intelligence, 27, 141-156.

Chaiken, S. R. (1993). Two models for an inspection time paradigm: Processing distraction

and processing speed versus processing speed and asymptotic strength. Intelligence,

17, 257-283.

Chaiken, S. R., & Young, R. K. (1993). Inspection time and intelligence: Attempts to

eliminate the apparent movement strategy. American Journal of Psychology, 106,

191-210.

Crawford, J. R., Deary, I. J., Allan, K. M., & Gustafsson, J. E. (1998). Evaluating competing

models of the relationship between inspection time and psychometric intelligence.

Intelligence, 26, 27-42.

Perceptual learning on inspection time 21

Deary, I. J. (2000). Looking down on human intelligence: From psychometrics to the brain.

Oxford: OUP.

Dorais, A., & Sagi, D. (1997). Contrast masking effects change with practice. Vision

Research, 37, 1725-1733.

Dosher, B. A., & Lu, Z. L. (1999). Mechanisms of perceptual learning. Vision Research, 39,

3197-3221.

Evans, G., & Nettelbeck, T. (1993). Inspection time: A flash mask to reduce apparent

movement effects. Personality and Individual Differences, 15, 91-94.

Eysenck, H. J. (Ed.) (1982). A model for intelligence. NY: Springer-Verlag.

Fahle, M. & Poggio, T. (Eds.) (2002). Perceptual learning. Cambridge, MA: The MIT Press.

Fiorentini, A., & Berardi, N. (1980). Perceptual learning specific for orientation and spatial

frequency. Nature, 287, 43-44.

Garlick, D. (2002). Understanding the nature of the general factor of intelligence: The role of

individual differences in neural plasticity as an explanatory mechanism.

Psychological Review, 109, 116-136.

Howell, D.C. (2002). Statistical methods for psychology (5th ed.). Pacific Grove CA:

Duxbury.

Karni, A., & Sagi, D. (1993). The time course of learning a visual skill. Nature, 365, 250-

252.

Knibb, K. (1992). A dynamic mask for inspection time. Personality and Individual

Differences, 13, 237-248.

Mackintosh, N. J., & Bennett, E. S. (2002). IT, IQ and perceptual speed. Personality and

Individual Differences, 32, 685-693.

Nettelbeck, T. (1987). Inspection time and intelligence. In P.A. Vernon (Ed.), Speed of

information processing and intelligence (pp. 295-346). Norwood, NJ: Ablex.

Perceptual learning on inspection time 22

Nettelbeck, T. & Vita, P. (1992). Inspection time in two childhood age cohorts: A constant or

a developmental function? British Journal of Developmental Psychology, 10, 189-

197.

Nettelbeck, T. & Wilson, C. (1985). A cross-sequential analysis of developmental differences

in speed of visual information processing. Journal of Experimental Child Psychology,

40, 1-22.

Nettelbeck, T., Rabbitt, P. M. A., Wilson, C., & Batt, R. (1996). Uncoupling learning from

initial recall: The relationship between speed and memory deficits in old age. British

Journal of Psychology, 87, 593-607.

Poggio, T., Fahle, M., & Edelman, S. (1992). Fast perceptual learning in visual hyperacuity.

Science, 256, 1018-1021.

Ramachandran, V. S., & Braddick, O. (1973). Orientation-specific learning in stereopsis.

Perception, 2, 371-376.

Ramachandran, V. S., & Hirstein, W. (1998). The perception of phantom limbs: The D. O.

Hebb lecture. Brain, 121, 1603-1630.

Sadler, A. J., & Deary, I. J. (1996). Cerebral asymmetries in inspection time?

Neuropsychologia, 34, 283-295.

Simonton, D. K. (2003). Genius and g: Intelligence and exceptional achievement. In H.

Nyborg (Ed.), The scientific study of general intelligence: Tribute to Arthur R Jensen

(pp. 229-245). Kidlington, Oxford: Pergamon.

Spearman, C. (1927). The abilities of man: Oxford: Macmillan.

Stokes, T. L., & Bors, D. A. (2001). The development of a same-different inspection time

paradigm and the effects of practice. Intelligence, 29, 247-261.

Perceptual learning on inspection time 23

Stough, C., Bates, T. C., Mangan, G. L., & Colrain, I. (2001). Inspection time and

intelligence: Further attempts to eliminate the apparent movement strategy.

Intelligence, 29, 219-230.

Vernon, P. A. (Ed.). (1993). Biological approaches to the study of human intelligence.

Norwood, NJ: Ablex.

Vickers, D., Nettelbeck, T., & Willson, R. J. (1972). Perceptual indices of performance: The

measurement of "inspection time" and "noise" in the visual system. Perception, 1,

263-295.

Wetherill, G. B., & Levitt, H. (1965). Sequential estimation of points on a psychometric

function. British Journal of Mathematical and Statistical Psychology, 18, 1-10.

White, M. (1996). Interpreting inspection time as a measure of the speed of sensory

processing. Personality and Individual Differences, 20, 351-363.

Woodcock, R. W., & Johnson, M. B. (1989). Woodcock-Johnson Psycho-Educational

Battery-Revised. Allen, TX: DLM Teaching Resources.

Zanker, J. M. (1999). Perceptual learning in primary and secondary motion vision. Vision

Research, 39, 1293-1304.

Zohary, E., Celebrini, S., Britten, K. H., & Newsome, W. T. (1994). Neuronal plasticity that

underlies improvement in perceptual performance. Science, 263, 1289-1292.

Perceptual learning on inspection time 24

Author Note

Nicholas R Burns, Ted Nettelbeck and Jason McPherson, Department of Psychology,

University of Adelaide, Australia. Lazar Stankov, University of Sydney, Australia.

Lazar Stankov is now at Center for New Constructs, Educational Testing Services,

Princeton, NJ.

This research was supported by a Faculty of Health Sciences B3 Grant awarded to the

first author. Preparation of the manuscript was supported by an Australian Research Council

Discovery Grant DP0211113 awarded to the first two authors. We thank Dr Nathan Brody

and Dr Timothy Bates for helpful comments on earlier versions of this paper.

Correspondence concerning this article should be addressed to Nicholas Burns,

Department of Psychology, University of Adelaide, South Australia, 5005, Australia. E-mail:

nicholas.burns@adelaide.edu.au

Perceptual learning on inspection time 25

Table 1

Means and 95% confidence intervals for threshold estimates averaged across the

first two (Initial) and last two (Final) staircase estimates from the training phase and the two

staircase estimates from the test phase (Test)for Vertical IT, Horizontal IT, Vertical Motion

and Horizontal Motion tasks

Vertical IT

(msec)

Horizontal IT

(msec)

Vertical Motion

(% noise)

Horizontal Motion

(%noise)

Mean CI95 Mean CI95 Mean CI95 Mean CI95

Initial

a

51.4 44.9, 57.9 64.8 57.8, 71.8 .54 .51, .56 .51 .48, .55

Finalb 40.8 34.2, 47.4 41.6 34.6, 48.6 .59 .57, .61 .61 .58, .63

Testc 55.6 48.1, 63.0 46.1 39.2, 53.0 .54 .52, .57 .56 .53, .59

a Initial is the average of the first two threshold estimates of the training phase for

each task

b Final is the average of the last two threshold estimates of the training phase for

each task

c Test is the average of the two threshold estimates of the test phase for each task.

Thus participants trained on Vertical IT were tested on Horizontal IT, and so on

Perceptual learning on inspection time 26

Table 2

Means and standard deviations for cognitive ability test scores for 56 participants

and correlations, with 95 confidence limits, of tests and principal component scores with

improvement on psychophysical tasks

Improvement IT Improvement Motion

WJ–R a Mean SD r CI95 r CI95

Visual Matching 539.5 13.1 .20 [-.07, .44] .06 [-.21, .32]

Cross Out 528.6 8.4 .31 [.05, .53] -.18 [-.42, .09]

Analysis-Synthesis 523.9 9.7 -.20 [-.44, .07] .02 [-.24, .28]

Concept Formation 527.9 11.6 .14 [-.13, .39] .10 [-.17, .35]

STOCA b

Line Length 69.8 12.7 .16 [-.11, .41] -.22 [-.46, .05]

Letter Spotting 58.2 11.0 .30 [.04, .52] .08 [-.19, .34]

Principal Component Scores c

Fluid Ability (Gf) - - .10 [-.17, .35] -.02 [-.29, .24]

Speed of Processing (Gs) - - .35 [.10, .56] -.06 [-.32 .21]

a Scores on tests from Woodcock-Johnson Psycho-Educational Battery – Revised are

Rasch-scaled W-scores centered on 500

b Scores on tests from Stankov’s Tests of Cognitive Ability are percentage items

correct

c Scores for obliquely rotated principal components. Gf has loading from Analysis-

Synthesis and Concept Formation; other tests load on Gs

Perceptual learning on inspection time 27

Table 3

Correlations of Vertical IT and Horizontal IT with Gs and Gf at each staircase

estimate

Staircase Estimate Vertical IT Horizontal IT

Gs Gf Gs Gf

1 -0.250 -0.370 0.142 -0.009

2 -0.349 -0.282 0.119 -0.021

3 -0.526 -0.415 0.037 0.099

4 -0.268 -0.087 -0.243 -0.091

5 -0.411 -0.201 -0.130 -0.269

6 -0.459 -0.325 -0.160 0.008

7 -0.342 -0.283 -0.046 0.079

8 -0.396 -0.271 0.067 0.059

9 -0.298 -0.155 -0.074 0.024

10 -0.279 -0.135 -0.008 0.019

11 -0.300 -0.129 0.173 0.278

12 -0.463 -0.322 0.128 0.203

13 -0.471 -0.302 0.198 0.172

14 -0.564 -0.163 -0.019 -0.020

15 -0.443 -0.155 -0.287 -0.101

16 -0.451 -0.089 -0.154 0.059

17 -0.346 -0.284 -0.220 -0.214

18 -0.294 -0.169 -0.192 -0.132

Note. An absolute correlation of .37 is statistically significant at p = .05 level for N =

28 in each group.

Perceptual learning on inspection time 28

Figure Captions

Figure 1. (Top) Two alternative IT target figures (center), and lightning mask (right) for the

inspection time (IT) task. Shown is the Vertical IT task; the Horizontal IT task used identical

stimuli rotated 90 deg anticlockwise. (Bottom) Random dot kinematogram (RDK) stimulus at

one instant of time. The dots within the white square are displaced coherently in the direction

of the white arrow from frame-to-frame. The background dots are black or white with equal

probability in any one frame of the animation and their state is uncorrelated from frame-to-

frame. Note that the black and white borders of the square and background do not appear in

the actual stimuli, the object is defined by its motion against the dynamic noise background

which is presented on a background of the same mean luminance as the RDK.

Figure 2. (Top) Mean 79% estimates for threshold SOA for Vertical IT and Horizontal IT

(msec) at each staircase estimate. Note, the legend refers to training groups (estimates 1-16);

at estimates 17 and 18 each group’s legend remains the same but participants have completed

the orthogonal version of the task. (Bottom) Mean 79% threshold estimates for Vertical

Motion and Horizontal Motion (% noise) at each staircase estimate. As for IT, above, the

legend refers to training groups and at estimates 17 and 18 each group completed the

orthogonal task. Error bars are SEM.

Perceptual learning on inspection time 29

Figure 1 Burns et al.

Perceptual learning on inspection time 30

0

10

20

30

40

50

60

70

80

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Sequence of estimates

IT (m

sec)

Vertical ITHorizontal IT

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Sequence of estimates

Noi

se (%

)

Vertical MotionHorizontal Motion

Figure 2 Burns et al.