Attention and Perceptual Learning Modulate Contextual Influences on Visual Perception

Transcription

1 Neuron, Vol. 20, , June, 1998, Copyright 1998 by Cell Press Attention and Perceptual Learning Modulate Contextual Influences on Visual Perception Minami Ito, Gerald Westheimer, and Charles D. Gilbert* The Rockefeller University New York, New York Summary Yet another factor influencing the ability to discriminate visual attributes is perceptual learning. A wide variety of segmentation, orientation, and hyperacuity tasks improve with practice. Though there are also instances of specificity to a particular stimulus configuration, implying a top-down influence (Shiu and Pashler, 1992; Treisman et al., 1992; Ahissar and Hochstein, 1993; Fahle and Morgan, 1996; Crist et al., 1997), many examples of perceptual learning demonstrate the specificity to stimulus location and to such parameters as orientation and movement that suggest the involvement of early stages of visual processing (Gilbert, 1994; Sagi and Brightness discrimination thresholds and facilitation by lateral interaction were measured in five human observers and two monkeys. The subjects judged the brightness of one of four peripherally seen lines against Tanne, 1994; Crist et al., 1997). a reference. This experiment was performed both We are here studying the interaction between these when the observer was cued to the position of the test three modulations of early visual processing: context, line (focused attention) and when there was no cue attention, and perceptual learning. We analyze how the (distributed attention). Discrimination was better with brightness discrimination of a peripherally seen line is focused than with distributed attention. When the test affected first by the presence of a flanking line and then line had a collinear flank, its brightness was enhanced; by the attentional state. We further go on to show that this enhancement was four times more prominent with there is an interaction between these two influences distributed than with focused attention. After training, flank facilitation is much more marked with distributed thresholds improved and collinear facilitation decreased attention than focused attention. Finally, we examine under distributed but not under focused attention. The how these effects in turn change over time with practice findings show that there are fewer benefits from con- in the visual task. textual interaction once attention is directed toward a visual location, and that the attentional effects are subject to training. Results In our configuration, four lines were located radially like Introduction spokes around a central fixation point (Figure 1B) next to which there was a comparison line. In a given presen- Even in the perception of simple visual attributes, global tation, three of the spokes had the same intensity as characteristics of contours and surfaces as well as the the comparison line, and the fourth would have a higher animal s attentional state play a role. The saliency of a or lower intensity at random. The observer had to make contour that is, its ability to stand out in a complex a judgment of brighter or dimmer, and the result of visual environment depends, for example, on the geo- many such trials allowed the evaluation of two response metrical relationships of the line segments forming it parameters, (1) the threshold of brightness discrimina- (Wertheimer, 1938; Ullman, 1990; Field et al., 1993). tion (represented inversely by the slope of a psychomet- In an attempt to analyze the neurophysiological basis ric curve; Figure 2) and (2) a mean value, which is the of such contextual effects, it has been demonstrated test line intensity that matches that of the comparison that a line can be more easily detected when a collinear line and which is represented by the 50% point of the line is placed near it. The magnitude of this facilitation psychometric curves in Figure 2. In any one trial, the depends on the collinear separation, lateral offset, and line that was to be the test line was chosen at random, relative orientation of the lines (Dresp, 1993; Polat and but in one set of experiments the observer was cued as Sagi, 1993, 1994; Kapadia et al., 1995). These findings to which of the four spokes was to be the test line in matched the stimulus interactions in the response of the immediately following trial (focused attention), and cells in area V1 of the alert macaque (Kapadia et al., in another the observer was not cued and therefore had 1995). to distribute attention over all four spokes. As seen in Another powerful influence on an observer s ability to Figure 2, the observers had a lower brightness discrimidiscriminate small differences in stimulus attributes is nation threshold that is, they were more sensitive to the state of attention. Expectation of, or selective at- test line brightness changes when they had received tention to, a particular object or location can enhance a cue than when they were not cued. This is in accord perceptual sensitivity over that for unattended locations with previous findings on detection (Cohn and Lasley, or when there is uncertainty about the location (Bashin- 1974), orientation discrimination (Lindblom and Westski and Bacharach, 1980; Downing, 1988; Nakayama heimer, 1992a), and stereoacuity (Lindblom and Westand Mackeben, 1989; Lindblom and Westheimer, 1992a, heimer, 1992b). 1992b). When each of the four spokes has a flank (Figure 1B, right), there is an additional phenomenon: the lines appear to be brighter. This brightness facilitation was * To whom correspondence should be addressed. quantified by finding the test line intensity at which it

2 Neuron 1192 Figure 2. Typical Results of Attentional Effects in One Human Subject Two psychometric curves were obtained with each attentional state, one without (square) and the other with (circle) contextual flanks. Coordinates show the percent of target line appears brighter than reference responses as a function of luminance level of the target line expressed as a ratio of the luminance of the reference. The value 1 refers to luminance of the reference line. Continuous lines indicate the best-fitting psychometric curves obtained by the probit method, which yielded two measures: the threshold from the slope of the curves and the shift from the difference in brightness match to the reference. A shift of the psychometric curves toward the left indicates facilitation due to the flanking lines. Data demonstrate that the brightness discrimination is better (threshold is lower) and the contextual influence is less (shift is smaller) with focal attention than with distributed attention. over all four spokes, the shift was considerably larger; in other words, the brightness facilitation due to flanks was much more pronounced. These effects were measured on seven observers, five human subjects, and two monkeys and shown in Figures 3A and 3B. The average results, in Figures 3C and 3D, reveal that when attention has to be distributed over four lines instead of focused on just one line, bright- ness discrimination thresholds are decreased by more than a factor of two, whereas there is an almost 4-fold increase in facilitation. Figure 1. Experimental Design (A) Stimuli and sequence of presentation with a standard four position stimulus. Lines were bright against a dark background. The subject fixated the central spot (fixation point) and had to judge whether the target line was brighter or dimmer than the reference. In the focal attention trials, the observer was cued on which of the four peripheral lines was the target line during that exposure; in the distributed attention trials, any of the four could be the target. The diagram illustrates the sequence of events for a trial. In each trial, several stimuli were presentedafter one cue presentation, and subjects reported on each stimulus. The number of repetitions was changed in random fashion; the figure describes two stimulus presentations. (B) Appearance of the screen during stimulus presentation. (Left) Basic pattern of fixation point, reference line, and four periph- eral test lines, one of which would be the target. (Right) To examine the contextual effect, there was a set of somewhat brighter flanking lines in addition to a basic pattern. (C) Mean and standard deviations in the horizontal and vertical meridians of eye positions during stimulus presentation in a primate subject (UM, left) and a human subject (MI, right). Position was measured during stimulus presentation time (100 ms) while subjects performed the brightness discrimination task. The means and standard deviations were calculated during a 1 day session for each horizontal and vertical direction and for each cued location. In each attentional condition, the mean eye position was calculated for each target location. Data points overlapped between focal attention (dia- monds) and distributed attention (squares). The small circles in the four corners indicate the locations of test lines. matches the comparison; it is less than when there is no flank. This is the effect described and analyzed by Kapadia et al. (1995) and is represented in the plots of Figure 2 by a leftward shift of the psychometric curve obtained with a flank with respect to that obtained without a flank. The magnitude of this shift is, however, quite different in the two attentional states. When the subject had not been cued and hence had to distribute attention Influence of Training on Attentional Effects To study the effect of practice, our subjects performed the brightness discrimination task both with and without contextual flanks and under both the focused and the distributed attention regimens for several months. Ses- sions, each with a total of 560 trials, were conducted 3 5 times a week for 4 6 months. The training periods were 18 (subject KI), 19 (PN), 28 (MI), 22 (MK), 20 (AG), 13 (SA), and 19 (UM) weeks for the seven subjects. Psychometric curves were evaluated weekly for each subject and each experimental condition. Learning and Brightness Discrimination Typical results, plotted in Figure 4A for a human subject and in Figure 4B for a monkey, reveal that there is a steady decrease in the brightness discrimination thresh- old with training in the distributed attention condition. But in the focused attention conditions there is little change, so that the two curves converge. Normalized average learning effects in all subjects for the two attentional conditions are shown in Figure 4C. Learning and Contextual Effects As can be seen in the individual curve for observer PN (Figure 5A), the facilitatory effect of flanks on the apparent brightness of the test lines decreased with training when there was distributed attention. In the case of

3 Attention and Perceptual Learning 1193 Figure 3. Effect of Attention on Brightness Thresholds and Contextual Facilitation Data for our seven subjects, including monkeys (SA and UM), for brightness discrimination (A and C) and contextual facilitation (B and D) from results such as those shown in Figure 2. (A and B) Gray and black bars refer, respectively, to the distributed and focal attention regimens. Error bars and levels of significance (*p 0.05, **p 0.01) are shown above each set of paired conditions. Attentional modulation was consistent among our subjects. (C and D) Averaged data for all seven observers. Brightness discrimination thresholds, both with and without contextual flanks, are better by a factor of about two when a previous cue instructs the observer of the expected position of the target than when all four possible target locations are equally likely. On the other hand, the flanks have a much higher influence on the target brightness when the attention is distributed. to appear in the subsequent test period, improved the accuracy of perceptual judgements. The performance of all subjects was poor under distributed attention, suggesting that when a limited resource of attention is divided among many possible target locations, the benefit of focal attention is reduced or disappears. It is known that sensitivity can be manipulated by spread of attention (Beck and Ambler, 1973; Cohn and Lasley, 1974; LaBerge, 1983; Eriksen and St. James, 1986; Krose and Julesz, 1989; Lindblom and Westheimer, 1992a, 1992b; Balz and Hock, 1997). Thresholds are lower with narrowly focused attention than with broadly spread attention. The identification of the odd stimulus in our task demanded serial search, though the brief exposure did not allow enough time for that. It has been estimated that at least 50 ms are needed for each location when attention is focused sequentially across all targets (Ber- gen and Julesz, 1983; reviewed by Egeth and Yantis, 1997). Since our pattern consisted of four test lines and the presentation time was 100 ms, not enough time would be available for serial search in our task. A salient finding in our study is the role of attention in modulating contextual effects. Unless there had been specific training in a given configuration, the facilitatory influences of a flanking line are much greater under distributed attention than under narrowly focused attention. In natural visual scenes, certain stimulus character- istics endow components of the image with saliency, enabling those components to emerge from the back- ground and attract attention. When attention is already focused attention, the effect was initially smaller (Figure 2) and did not change much with practice so that the two curves converged, as was the case for thresholds. Summary results for all of our seven subjects are given in Figure 5B. The results for the monkeys were very similar to those seen in human observers. Specificity of Learning To examine whether the training described above was specific to the stimulus location, we obtained measurements for brightness discrimination thresholds for a configuration identical to that used so far, except that it was rotated through 45 so that the four lines were along the 90 and 180 meridians. Data were obtained for the diffuse attention situation both during the second week of training in the previous task and during the last week of training in three observers. These measurements were carried out in separate blocks for the ones used in assessing the effect of training on the 45 /135 lines. It was found (Figure 6) that there was a substantial transfer of the improvement in the brightness discrimination thresholds from the retinal locations that were used for training to ones 45 away. Discussion In the present study, we used a brightness discrimination task as a probe to study the effects of attention on visual perception. The subjects performance was clearly dependent on the status of attention; advance information about location, that is, where the target was

4 Neuron 1194 focused on a particular component, however, there is no further saliency derived from contextual effects. There appears to be an interchangeability between the factors that endow a visual pattern with saliency and the directing of attention toward that pattern by an independent cue. Either process causes the pattern to stand out from a complex background, and once attention is directed toward an object, the saliency derived by context is less. The contextual facilitation that still remains under focal attention would have a value in mediating contour integration for attended stimuli. It might be argued that any distributed attention task must ultimately involve a component of focal attention in that, in order to make a discrimination about a specific object, an observer has to pay attention to that object. But our data show a difference in performance under our two attentional regimens, and hence they are not equivalent. We find a substantial difference in flank facilitation measured under focal and distributed attention and at this stage cannot make the attempt to parse the stages involved in a distributed attention task. In this connection, it is of interest that the performance under distributed attention is different from that under focal attention regardless of the time available to bring one s attentional focus to the target (Lindblom and Westheimer, 1992a). Under focal attention, the subject is Figure 5. Influence of Training on Contextual Facilitation Figure 4. Influence of Training on Brightness Thresholds Time course of theimprovement of thethreshold for a human subject (A) and a monkey (B). Threshold was measured weekly under distrib- uted attention (square) and focal attention (circle) during the training period. Error bars show the standard error of the mean yielded by the probit method. (C) shows the threshold averaged among our seven subjects, including monkeys. The training reduced the bright- ness discrimination threshold by a factor of about two when all four possible target locations are equally likely. On the other hand, improvement was limited when a previous cue instructed the ob- server as to the expected position of the target. (A) Amplitude of facilitation plotted in each weekly session under distributed attention (square) and focal attention (circle). Initially, facilitation is larger under distributed attention than focal attention. Training reduced facilitation under distributed attention, thereby eliminating the difference observed initially between distributed and focal attention. (B) Results averaged for all seven observers. The training reduced the effect of facilitation by a factor of about three when the observer had to distribute his attention to all four possible target locations. cued to the target location before the target appears and therefore does not depend on the target attributes to attract attention to that location. On the other hand, Figure 6. Transfer of Training Effects to Untrained Locations The threshold of the brightness discrimination was measured with the untrainedpositions stimulus only on the second andlast weeks of the training period. The untrained stimulus was given by rotating the original four position stimulus by 45. The averaged result from three human subjects (KI, PN, and AG) showed similar reduction of the threshold in the untrained positions as in the trained positions.

5 Attention and Perceptual Learning 1195 when attention is distributed, the stimulus character is Experimental Procedures centrally important in identifying the target, and then Brightness discrimination was studied in five human observers (male the flank interaction is seen to help a lot. and female, years of age) with normal or corrected-to-normal The learning that we observedunder distributed atten- vision and two Rhesus monkeys (Macaca mulatta, adult males, tion suggests that observers can increase the number weighing 4.4 and 4.8 kg). All procedures followed recommendations of attentional foci maintained in parallel. This is perhaps from the declaration of Helsinki and NIH guidelines on the care and analogous to the idea that serial search tasks can be use of laboratory animals. The basic experimental paradigm involved the judgment by the transformed into parallel search tasks by perceptual observer whether a peripherally seen test line appeared brighter or learning (Sireteanu and Rettenbach, 1995). In our experidimmer than a centrally presented comparison line. During a given ments, the subjects performance depended on the trial, the test line was shown with one of seven equally spaced number of stimuli, and due to the limitation in attentional luminous intensity steps, ranging from three modules dimmer than resources, the distribution of attention occurred at the the comparison line through equality to three modules brighter cost of sensitivity to visual attributes. Our results sugwhether than the comparison line. The observer had to respond merely the test line appeared brighter or dimmer. A run of such gest that distributed attention, ratherthan being diffused presentations, each with a test line intensity picked at random from over a large area, can involve multi-focal attention to this ensemble of seven, allowed the accumulation of a psychometric a number of discrete locations. The specificity of the curve (Figure 1). It has the shape of an ogive, demonstrating an learning indicates that the improvement is not simply increasing percentage of yes responses to the question was the one of maintaining an increased number of attentional test line brighter than the comparison? with increasing test line intensity. Using the standard probit technique (Finney, 1952), a cu- searchlights but is connected to a particular stimulus mulative Gaussian curve was fitted to these data. The analysis gives configuration. The improvement with training is seen for two parameters, each with its own standard error: the 50% point, the four member array, whether the stimuli are shown which indicates the test line intensity at which the test and compariat the 45 /135 meridia or the 0 /90 meridia. Conceiv- son appear to match brightness, and the slope, which is an indication ably, the subjects may generate an attentional profile of the observer s sensitivity to brightness changes. The shalably, that matches the four member template and may be capato lower the curve that is, the more the physical test line intensity has be varied for the observer to manifest a response difference the ble of performing a rapid mental rotation of that template poorer the brightness discrimination or, in other words, the higher as long as there are no other distractors present. the brightness discrimination threshold. This analysis was per- Because the contextual effects that we have studied formed on each response curve and a chi-square test generally psychophysically are reflected in the response proper- revealed satisfactory conformity to a Gaussian template. A change ties of cells in area V1 (Kapadia et al., 1995), it is tempting in perceived brightness of the test line is seen as a left or right shift to speculate that the attentional modulation of these in the position of the probit curve and is read as the brightness at which the subject is performing at 50%. effects might also be exerted in V1. The specificity of the The pattern of stimulus lines was designed to permit testing in learning effect for stimulus configuration rather than for the presence or absence of a test line flank and also with two stimulus position, however, suggests a possible interac- different attentional states. There was a centralfixation point, slightly tion between V1 and higher order visual cortical areas, lateral to the comparison line to provide the brightness criterion, where cells are specific for shape and not for position. and four stimulus lines at 4 eccentricity (3.5 for the monkeys) set radially along the 45 and 135 meridians. Each stimulus line could Our results are consistent with the idea that top-down have a companion flank (Figure 2). modulation of high level attentional mechanisms is cru- The basic brightness-match judgments were performed under cial for perceptual learning (Shiu and Pashler, 1992; two attention regimens: (1) focused attention, in which one of the Treisman et al., 1992; Ahissar and Hochstein, 1993; Fahle four peripheral stimulus lines was the test line whose brightness and Morgan, 1996; Crist et al., 1997). The learning that was changed and the observer was cued beforehand as to which we have observed in the domain of attention is somefour one it was, and (2) distributed attention, in which again one of the peripheral lines had the brightness change, but the observer what different in character from that observed in percepwas not informed of which one it was and, therefore, had to distribute tual learning of stimulus attributes, which is much more attention to all four. In each of these situations, the brightnessspecific for stimulus location (Schoups et al., 1995; Ahis- match experiments were performed both in the presence and in the sar and Hochstein, 1996; Crist et al., 1997). absence of a flanking line. The similarity in our findings between the human and the macaque (SA and UM) subjects raises confidence Monkey Subjects The macaques were trained in a task similar to that described above in using the macaque for physiological studies of the except that they reported on only the last stimulus in a cycle by mechanisms of human perception. The primary differmaking a saccade to one of two saccade targets, which were given ence between the two species is the smaller difference after the fixation point disappeared. The training protocol was that between distributed and focal attention in the macaque described earlier (Kapadia et al., 1995). Each trial was initiated when than in the human. This can be attributed to training the animal pulled the lever attached to the apparatus. Animals were effects, since, as we have shown, the differences seen on a regimen that restricted their water intake 6 days a week, and a dropof juice was given as a reward when they performed appropriunder different states of attention are reduced with trainately in the task. The monkeys were trained by gradually modifying ing. The macaques had already received a measure of their task from simple fixation to the final form of the discrimination training before psychometric curves could be obtained. task. The apparatus for the psychophysical studies on the monkeys The comparable results obtained for human and macd/m presented test and flanking lines (18 4 ) at a luminance of 37.7 caque support previous reports on the similarity of the 2 against a background of 17.7 cd/m 2. The experiments were conducted in a darkened room, binocularly two species (De Valois et al., 1974a, 1974b; Teller, 1981; with natural pupils. Human observers used a chin and forehead rest Vogels and Orban, 1990). Therefore, primate subjects to ensure head stability. Stimuli were presented on a cm provide an opportunity to explore the physiological sub- CRT screen (NEC Multisync 6FG) viewed at 117 cm. Lines were 30 strates of attentional effects. arcmin long and 5 arcmin wide with a luminance of 20.8 cd/m 2

6 Neuron 1196 against a uniform background of 8.5 cd/m 2. Five hundred millisec- References onds after the initial cue, the stimulus was exposed for 100 ms and there was a 900 ms pause allowing the subject to record a response Ahissar, M., and Hochstein, S. (1993). Attentional control of early before the next stimulus cycle. To test for contextual effects, each perceptual learning. Proc. Natl. Acad. Sci. USA 90, test line was accompanied by a 30 5 flanking line, with a lumi- Ahissar, M., and Hochstein, S. (1996). Learning pop-out detection: nance of 32 cd/m 2, of the same orientation and situated 1 further specificities to stimulus characteristics. Vision Res. 36, out from the fixation point along its axis of orientation. Bach, M., Bouis, D., and Fischer, B. (1983). An accurate and linear One of the inner set of spoke lines was the test line. The others, infrared oculometer. J. Neurosci. Methods 9, as well as the comparison line, all remained at a luminance of 20.8 cd/m 2. The test line was shown at one of a range of seven lumion spatial resolution. Vision Res. 37, Balz, G.Y., and Hock, H.S. (1997). The effect of attentional spread nances, centered on 20.8 cd/m 2. The flanks (outer ring of spokes) had a luminance of 32 cd/m 2. The range of test line luminances used Bashinski, H.S., and Bacharach, V.R. (1980). Enhancement of per- depended on the subject s sensitivity, but on almost all occasions ceptual sensitivity as the result of selectively attending to spatial the test line remainedbelow the flank in luminance. In the distributed locations. Percept. Psychophys. 28, attention situation, the direction of luminance difference between Beck, J., and Ambler, B. (1973). The effects of concentrated and flank and test line could not generally be used as a cue: all of the distributed attention on peripheral acuity. Percept. Psychophys. 14, inner lines were dimmer than the flanks Data collected over a period of 1 week ( trials) yielded a Bergen, J.R., and Julesz, B. (1983). Parallel versus serial processing measure of the test line brightness that matched the reference, and in rapid pattern discrimination. Nature 303, also of the brightness discrimination sensitivity under the conditions of focused and distributed attention, in each case with and without Cohn, T.E., and Lasley, D.J. (1974). Detectability of a luminance contextual flanks. increment: effect of spatial uncertainty. J. Opt. Soc. Am. 64, 1715 When there is a peripheral stimulus and the observer knows where it will be shown, it is in theory possible for the observer to make a Crist, R.E., Kapadia, M.K., Westheimer, G., and Gilbert, C.D. (1997). predictive fixation shift to the new stimulus area and use the fovea Perceptual learning of spatial localization: specificity for orientation, instead of the periphery. We have several lines of evidence that this position and context. J. Neurophysiol. 78, was not a factor in our experiments. De Valois, R.L., Morgan, H.C., Polson, M.C., Mead, W.R., and Hull, First, the simulus was so short in duration that a saccade to the E.M. (1974a). Psychophysical studies of monkey vision. 1. Macaque stimulus would not be carried out during the exposure. Second, luminosity and color vision tests. Vision Res. 14, observers, even untrained ones in clinical eye-testing situations, can De Valois, R.L., Morgan, H.C., and Snodderly, D.M. (1974b). Psychoadequately hold central fixation even as they attend and respond to physical studies of monkey vision. 3. Spatial luminance contrast targets in the retinal periphery (Helmholtz, 1896; Eriksen and Hoffsensitivity tests of macaque and human observers. Vision Res. 14, man, 1973; Posner, 1980; LaBerge, 1983; Mackeben and Nakayama, ). Perimetric examination of the retinal periphery, for which quite naive patients must maintain centralfixation while reporting on Downing, C.J. (1988). Expectancy and visual-spatial attention: ef- peripheral stimuli, is a routine clinical procedure. Third, thresholds in fects on perceptual quality. J. Exp. Psychol. Hum. Percep. Perform. the fovea are many timesbetter than in the periphery, yet our observ- 14, ers consistently showed the high thresholds characteristic of the Dresp, B. (1993). Bright lines and edges facilitate the detection of periphery. Fourth, after training (Figure 4A), thresholds in the focused small light targets. Spat. Vision 7, and the distributed attention situations were essentially the Egeth, H.E., and Yantis, S. (1997). Visual attention: control, represen- same, yet only in the former was therea cue that would have enabled tation, and time course. Annu. Rev. Psychol. 48, the subject to move fixation. And fifth, to provide evidence for the Eriksen, C.W., and Hoffman, J.E. (1973). The extent of processing contention that the subjects fixation remained steady and central, of noise elements during selective encoding from visual displays. we recorded the eye positions during a typical focused attention Percept. Psychophys. 14, run (Figure 1C). Fixation remained centered throughout. We conclude, therefore, that fixation errors could not account for Eriksen, C.W., and St. James, J.D. (1986). Visual attention within differences between the focused and distributed attention situa- and around the field of focal attention: a zoom lens model. Percept. tions. Psychophys. 40, Fahle, M., and Morgan, M. (1996). No transfer of perceptual learning Eye Position Monitoring between similar stimuli in the same retinal position. Curr. Biol. 6, We recorded eye positions in the monkeys, whose heads were held steady by an implanted post, by the conventional search coil method Field, D.J., Hayes, A., and Hess, R.F. (1993). Contour integration by (CNC Engineering; Judge et al., 1980). When the animal s fixation the human visual system: evidence for a local association field. crossed the border of the fixation window (30 arcmin), the task was Vision Res. 33, aborted. The averaged eye position and its standard deviation for Finney, M.A. (1952). Probit Analysis (Cambridge: Cambridge Univera 1 day session are shown in Figure 1C (left). The eye positions sity Press). for the different attentional states were indistinguishable. We also measured eye positions with an infrared oculometer (Bouis, Ger- Gilbert, C.D. (1994). Early perceptual learning. Proc. Natl. Acad. Sci. many; Bach et al., 1983) in one human subject while he performed USA 91, peripheral luminancediscrimination (Figure 1C, right). Here, the subond Helmholtz, H.V. (1896). Handbuch der physiologischen Optik, Sec- ject s head was held steady by a chin-rest and bite bar. Although Edition (Hamburg: L. Voss.), pp there were relatively larger standarddeviations than with the eye coil Judge, S.L., Richmond, B., and Chu, F.C. (1980). Implantation of method in the monkey, presumably because the infrared monitoring magnetic search coils for measurement of eye position: an improved method is not fully immune to drifts and slight head movements, method. Vision Res. 20, we did not find any predictive shifts in the direction of the location Kapadia, M.K., Ito, M., Gilbert, C.D., and Westheimer, G. (1995). of the expected stimulus. Improvement in visual sensitivity by changes in local context: parallel Acknowledgments studies in human observers and in V1 of alert monkeys. Neuron 15, Krose, B.J., and Julesz, B. (1989). The control and speed of shifts This work was supported by a grant from the National Eye Institute. of attention. Vision Res. 29, We thank Kathleen McGowan, Andrew Glatz, and Kaare Christian for assistance. LaBerge, D. (1983). Spatial extent of attention to letters and words. J. Exp. Psychol. Hum. Percept. Perform. 9, Received November 17, 1997; revised April 4, Lindblom, B., and Westheimer, G. (1992a). Uncertainty effects in

7 Attention and Perceptual Learning 1197 orientation discrimination of foveally seen lines in human observers. J. Physiol. (Lond.) 454, 1 8. Lindblom, B., and Westheimer, G. (1992b). Spatial uncertainty in stereoacuity tests: implications for clinical vision test design. Acta Opthalmol. 70, Mackeben, M., and Nakayama, K. (1993). Express attentional shifts. Vision Res. 33, Nakayama, K., and Mackeben, M. (1989). Sustained and transient components of focal visual attention. Vision Res. 29, Polat, U., and Sagi, D. (1993). Lateral interactions between spatial channels: suppression and facilitation revealed by lateral masking experiments. Vision Res. 33, Polat, U., and Sagi, D. (1994). The architecture of perceptual spatial interactions. Vision Res. 34, Posner, M.I. (1980). Orienting of attention. Quart. J. Exp. Psychophys. 32, Sagi, D., and Tanne, D. (1994). Perceptual learning: learning to see. Curr. Opin. Neurobiol. 4, Schoups, A.A., Vogels, R., and Orban, G.A. (1995). Human perceptual learning in identifying the oblique orientation: retinotopy, orientation specificity and monocularity. J. Physiol. (Lond.) 483, Shiu, L.-P., and Pashler, H. (1992). Improvement in line orientation discrimination is retinally local but dependent on cognitive set. Percept. Psychophys. 52, Sireteanu, R., and Rettenbach, R. (1995). Perceptual learning in visual search: fast, enduring, but non-specific. Vision Res. 35, Teller, D.Y. (1981). The development of visual acuity in human and monkey infant. Trends Neurosci. 4, Treisman, A., Veira, A., and Hayes, A. (1992). Automaticity and preattentive processing. Am. J. Psychol. 105, Ullman, S. (1990). Three-dimensional object recognition. Cold Spring Harbor Symp. Quant. Biol. 55, Vogels, R., and Orban, G.A. (1990). How well do response changes of striate neurons signal difference in orientation: a study in the discriminating monkey. J. Neurosci. 10, Wertheimer, M. (1938). Laws of organization in perceptual forms (London: Harcourt, Brace, and Jovanovich).