Dynamics of attentional deployment during saccadic programming

Abstract

The dynamics of attentional deployment before saccade execution was studied with a dual-task paradigm. Observers made a horizontal saccade whose direction was indicated by a symbolic precue and had to discriminate the orientation of a Gabor patch displayed at different delays after the precue (but before saccade onset). The patch location relative to the saccadic target was indicated to observers before each block. Therefore, on each trial, observers were informed simultaneously about the respective absolute locations of the saccadic and perceptual targets. The main result is that orientational acuity improved over a period of 150–200 ms after the precue onset at the saccadic target location, where overall performance is best, and at distant locations. This effect is due to attentional factors rather than to an alerting effect. It is also dependent on the efficiency of the temporal masks displayed before and after the Gabor patches.

Introduction

There is now clear evidence that visual attention is involved in the programming of saccades to select the saccadic target, especially when it appears in a cluttered environment (Liversedge & Findlay, 2000). Psychophysical experiments, using a dual-task paradigm, have convincingly shown that a stimulus presented in the period preceding saccade execution is best processed when its location coincides with the saccadic target (Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler, Anderson, Dosher, & Blaser, 1995). Perceptual performance for a stimulus displayed away from the saccadic target depends on a tradeoff between the different amounts of attentional resources allocated respectively to the motor task—as measured by saccadic accuracy and latency—and to the perceptual task—as measured by discrimination performance. When priority is given to the saccadic task, perceptual and saccadic performances are inversely related. However, when both tasks are given equal weights, perceptual performance can be significantly improved without much affecting the saccadic task. Therefore, although programming a saccade requires a spatial shift of attention to the saccadic target, some attention can be diverted from the saccadic goal at almost no cost for saccadic latency and accuracy (Kowler et al., 1995). 

What remains unclear is the temporal course of the attentional deployment during the period preceding the saccade onset. Classic studies on covert attention (i.e., when observers are required to fixate a single location while attending elsewhere) show a gradual buildup of activity at a precued location (Cheal & Lyon, 1991a, 1991b; Kinchla, 1992; Murphy & Eriksen, 1987; Nakayama & Mackeben, 1989). However, it is still not known whether the enhancement of processing at the saccadic target location is an abrupt all-or-none switching or a gradual buildup. If covert attention and attentional selection of a saccadic target share common mechanisms (Findlay & Gilchrist, 2003; Kustov & Robinson, 1996; Moore, Armstrong, & Fallah, 2003; Nobre, Gitelman, Dias, & Mesulam, 2000; Rizzolatti, Riggio, Dascola, & Umilta, 1987), then we could expect similar temporal enhancement in the two cases. Another unresolved issue concerns the ability of the attentional system to share resources between different tasks and/or locations during saccadic programming. Some authors have argued that attentional resources are strictly locked at the saccadic target location as long as the saccade is not executed (Deubel & Schneider, 2003), a point that is, however, controversial (Gersch, Kowler, & Dosher, 2004). Answering these questions should also help us understand the functional significance of enhanced presaccadic activity in low-level visual areas (Moore, Tolias, & Schiller, 1998; Super, Van Der Togt, Spekreijse, & Lamme, 2004). 

In the present study, we used a dual-task paradigm to assess the presaccadic dynamics of attentional deployment. Observers had to make a single horizontal saccade whose direction was indicated by an endogenous cue (primary task). They also had to report the tilt of a Gabor patch briefly presented at different delays after the cue but before the onset of saccadic eye movement (secondary task). Orientation discrimination is a well-established task that allows accurate assessment of the effects of attentional modulations resulting from endogenous or exogenous cuing (Morgan & Solomon, 2005; Pestilli & Carrasco, 2005; Solomon, 2004). Our main finding is that orientational acuity rapidly improved within 150–200 ms from the onset of the saccadic cue: Interestingly, this improvement over time occurs not only at the saccadic target location but also at distant locations, although overall performance is worse in the latter case. 

Materials and methods

Stimuli and procedures

The sequence of stimuli and the experimental task are shown in Figure 1. Observers fixated a dot displayed at the center of a circular array of eight disks (eccentricity = 6 deg). After a fixed delay (1,500 ms), the fixation point was replaced by an arrow (saccadic cue) indicating the direction of the required horizontal saccade so that the saccadic target was either the left or the right disk. The duration of the arrow was 6.25 ms (one frame). Disappearance of the arrow constituted the go signal to make the saccade. After a variable delay, all disks were replaced with Gabor patches (seven vertical, one tilted) for 50 ms (eight frames). As shown in Figure 2, the tilted patch could be presented in one of three relative locations with respect to the saccadic target: same (relative angle = 0 deg), above (+45 deg), or below (−45 deg). The relative location of the tilted patch with respect to the saccadic target was constant across a block and was indicated to observers at the beginning of each block. The only exception was in Experiment 3, where the tilted patch could appear at any of the eight disks' locations. 

Figure 1

 

Temporal sequence of stimuli when the saccadic target and the tilted patch coincide. Observers fixate the center of a circular array of eight disks. The fixation point is then replaced by a saccadic cue pointing either leftward or rightward. The offset of the cue is the go signal to execute the saccade. In this example, a leftward saccade has to be made to the left disk. After a variable delay, all disks are replaced by vertical Gabor patches except for one of them that is slightly tilted. Here, the tilted patch appears at the location of the saccadic target (as indicated by the dashed white circle, which was not present in the actual experiments). At the beginning of each experimental block, observers were informed of the relative position of the tilted patch with respect to the saccadic target (see Figure 2).

Figure 2

 

Relative angle of tilted patch location with respect to the saccadic target. In a given experimental block, the tilted patch could be located at one of three relative locations with respect to the saccadic target: it could be above the saccadic target at a relative angle of +45 deg, be below the saccadic target at an angle of −45 deg, or coincide with the saccadic target at a relative angle of 0 deg. Because the saccades were always horizontal, the tilted patch appeared in the upper visual field in the +45-deg condition and in the lower visual field in the −45-deg condition.

A mask consisting of static noise dots covering the whole screen immediately followed the Gabor patches for 37.5 ms (six frames). Finally, the initial array of eight noise disks was displayed until observers reported the orientation of the tilt with respect to vertical by pressing one of two buttons. 

The background was gray (∼20 cd/m²). The disks (∅ = 2 deg) and the whole screen mask consisted of white noise dots (0.1 × 0.1 deg) whose contrasts randomly ranged from 0% to ±50% relative to the mean luminance of the display (∼20 cd/m²). The only exception was in Experiment 4, where maximal contrast was 12.5%. Each of the Gabor patches was a 2-cycle/deg grating in cosine phase (contrast = 100%; mean luminance = ∼20 cd/m²) windowed by a Gaussian envelope (SD = 0.5 deg). 

The whole screen mask and the disks displayed after the patches were aimed at shortening the visible persistence of these patches (Coltheart, 1980b). Minimizing visible persistence is crucial if one wants to study the ability of the attentional system to be deployed across time. In other words, if persistence exceeds the physical duration of the patches too much, then it is difficult to ascertain whether attention was effective only around the period of physical display or during a much longer period including visible persistence (cf. Experiment 4). 

The importance of using a whole screen mask becomes apparent when considering the trials where the Gabor patches disappear just before a saccade's execution; in this case, if the Gabor patches were immediately followed by masking disks, then the following problem would arise: the retinal portion having processed a given patch just before the saccade would not be masked by the subsequently presented disk because of the rapid ocular displacement starting at the patches' disappearance. This problem is solved with a whole screen mask. 

Each experimental session usually consisted of three successive blocks corresponding to the three relative locations of the tilted Gabor patch. This was to ensure that different results obtained for these different conditions were based on comparable states (arousal, practice, etc.). 

Discarded trials

After each trial, saccade parameters were extracted to decide whether the trial was valid or not. To be considered as valid, a trial had to meet the two following criteria: (a) the horizontal amplitude of the saccade had to be within 20% of the required amplitude (6 deg). This is more conservative than most values used in previous relevant studies (e.g., 40% in Kowler et al., 1995). (b) The saccade onset had to occur after the Gabor patches' disappearance to make sure that perceptual processing of the patches was based on a presaccadic retinal image, which was entirely static (i.e., with no intrasaccadic smear). 

An auditory feedback was given when the first criterion was not met (i.e., when the actual saccade was too short or too long), thus helping observers keep a relatively constant accuracy during the course of a block (i.e., by giving a relatively constant weight to the attentional load implied in saccadic target selection). 

A trial was also aborted when the observer had not initiated an eye movement after a delay of 450 ms (from the saccadic cue disappearance). This time pressure was necessary to ensure that observers were engaged in a dual task; that is, saccadic programming occurred as soon as possible after the saccadic cue's disappearance, which is during the period when the patch was likely to appear. Otherwise, observers could have first focused their attention on the Gabor patch location (until the patch was displayed) and attentional selection of the saccadic target would have occurred only afterward. This in turn would have increased saccadic latency as already reported for trials where stimuli to be attended to did not coincide with the saccadic target (Kowler et al., 1995). 

Each trial considered as nonvalid was cancelled and another trial was initiated. A block was run until 50 (80 in the second set of experiments) valid trials were completed. 

Apparatus

Stimuli were displayed on a 21-in. color monitor (GDM-F520, Sony, Japan) driven by a display controller (Cambridge Research System VSG 2/3F, Cambridge, UK) with a 160-Hz frame rate (frame duration = 6.25 ms). At a viewing distance of 57 cm, the average separation between adjacent pixels subtended 0.046 deg of visual angle. The screen subtended 36 × 28 deg. A lookup table in the software was used to linearize the intensity response of the screen phosphors at an 8-bit luminance resolution. 

Threshold measurements

Orientation discrimination thresholds were measured by the method of single stimuli in which an observer had to decide whether the attended Gabor patch was tilted clockwise or counterclockwise relative to vertical. Each threshold was typically obtained by collecting between 250 and 500 valid trials (i.e., between 1,000 and 2,000 valid trials for each curve of four thresholds). 

In the first set of experiments, the tilted orientation could be randomly chosen between only two symmetrical orientations (e.g., −3 and +3 deg) within a given block. The particular pair of orientations used could be varied across blocks. We typically used six orientation values to sample a psychometric function. 

In the second set of experiments, the tilted orientation was chosen with adaptive staircase procedures (cf. details in the corresponding section). 

The data points, consisting of the measurements collected over all blocks, were fitted with a linear logistic model of the form to estimate the parameters of the psychometric function (Collett, 2003):  

$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(p_i\right)=\log [p_i/(1-p_i\left)\right]={\beta }_0+{\beta }_1{x}_i.$

(1)

 

The estimated parameters β₀ and β₁ were used to calculate the 0.75 and 0.5 thresholds:  

$\begin{array}{c}\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}{\mathrm{d}}_{0.75}=[\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}(0.75)-{\beta }_0]/{\beta }_1\\ \mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}{\mathrm{d}}_{0.5}=-{\beta }_0/{\beta }_1\end{array}.$

(2)

 

The orientation discrimination threshold was defined as the difference between the 0.75 and 0.5 thresholds:  

$\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{c}\mathrm{r}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{d}=\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(0.75\right)/{\beta }_1=g({\beta }_0,{\beta }_1).$

(3)

 

The standard error of thresholds was estimated with the following equation based on a standard result for the approximate variance of the bivariate function g (Collett, 2003, p. 108):  

${\left(\frac{\mathrm{\partial }g}{\mathrm{\partial }{\beta }_0}\right)}^2\mathrm{var}\left({\beta }_0\right)+{\left(\frac{\mathrm{\partial }g}{\mathrm{\partial }{\beta }_1}\right)}^2\mathrm{var}(\beta 1)+2\left(\frac{\mathrm{\partial }g}{\mathrm{\partial }{\beta }_0}\frac{\mathrm{\partial }g}{\mathrm{\partial }{\beta }_1}\right)\mathrm{cov}({\beta }_0,{\beta }_1),$

(4)

where var and cov, respectively, stand for variance and covariance. 

For each psychometric function, we measured the deviance statistic D to assess whether the fitted function displayed a lack of fit (Collett, 2003, p. 65). Values of D can be compared with tables of the χ² distributions with (n − p) degrees of freedom, where n is the number of orientations tested and p is the number of unknown parameters included in the logistic model (here p = 2). If the observed value of the D statistic exceeds the (100 × α)% point of the χ² distribution, where α is sufficiently small, then the lack of fit is declared significant at the (100 × α)% level. All thresholds reported (except three) rely on fitted functions for which no lack of fit was observed with α = .05. 

First set of experiments (Experiments 1 and 2)

Eye movement recording

Movements of the left eye were measured with an infrared video eye tracker (Iscan RK-716) at a sampling rate of 240 Hz. Acquisition of ocular data was triggered by a frame synchronization signal from the display controller to allow accurate alignment of eye movements and presentation of stimuli. Head movements were minimized with chin and head rests. 

Analysis of eye movement data

Calibrated eye position data were smoothed offline with a spline function to reduce noise and allow interpolation with millisecond accuracy (Busettini, Miles, & Schwarz, 1991). Eye velocity and acceleration were computed with a two-point differentiation. Horizontal saccades were detected using a 45-deg/s velocity threshold. Saccade onset and offset were defined, respectively, as 3 ms before and 3 ms after the velocity criterion was crossed. The saccade durations and peak velocities (not shown) calculated with our technique met the main sequence criteria (Collewijn, Erkelens, & Steinman, 1988; van der Geest & Frens, 2002). 

Observers

Two of the authors, with normal or corrected-to-normal vision, participated in the experiments. Although highly experienced in psychophysical observing, they were trained to perform Experiment 1 for approximately 10 hr to minimize perceptual learning over the period of the actual experiments. 

Stimuli

The delay between cue offset and appearance of the Gabor patches was randomly chosen, ranging from 6.25 to 200 ms by steps of 6.25 ms. 

Experiment 1a: Short duration of saccadic cue

The main goal of Experiment 1 was to establish whether perceptual performance would improve over time during the saccadic preparation period. The horizontal saccade was guided by a symbolic cue (a central arrow) that indicated which of the two disks displayed on the horizontal meridian was the saccade target (Figure 1). This saccadic cue was very short (one frame, 6.25 ms). Observers had to report the tilt (clockwise or counterclockwise with respect to vertical) of a Gabor patch briefly presented before the eye movement. Across an experimental block, the relative location of the tilted patch (e.g., 45 deg above the saccadic target) was constant. The Gabor patch was presented at different delays (from 6.25 to 200 ms by steps of 6.25 ms) after the saccadic cue's disappearance. We used this set of possible delays, rather than a few fixed delays, because we had no a priori hypothesis on the temporal scale of the effect and we wanted to be able to choose a posteriori the temporal bins that would be the most appropriate. 

Results

The delay between the offset of the saccadic cue and the onset of the Gabor patches was used to group the psychophysical data in successive 50-ms temporal bins. For each of the temporal bins, a psychometric function was fitted to the experimental points. Only three temporal bins are shown for observer SJ (vs. four bins for EC) because we were not able to collect enough data in the bin [150 ms, 200 ms[ (i.e., including 150 ms and excluding 200 ms) due to his shorter saccadic latencies (see Figure 4). 

Discrimination thresholds are plotted in Figure 3a as a function of the time elapsed since the saccadic cue's disappearance. After checking that no systematic difference between rightward and leftward saccades was observed, we averaged over the two possible saccadic directions (left and right). For both observers, the results clearly show an improvement of performance (i.e., a decrease in discrimination thresholds) over time. In all conditions, the 95% confidence intervals (not shown) for the first and the last bin never overlapped. The ratio of the discrimination thresholds corresponding to the first and the last bins is approximately 2 for each observer and relative location of the tilted patch. The lowest thresholds (∼1 deg for EC and ∼2 deg for SJ) were obtained in the last bin when the perceptual and the saccadic targets coincided. These absolute values are quite remarkable when considering that they were obtained in a dual-task paradigm and with short durations of the Gabor patches. Indeed, these values are in good agreement with those reported in studies on orientation acuity performed in much more optimal conditions, namely, with longer durations of the stimuli and most importantly with a single-task paradigm (Lee, Koch, & Braun, 1997; Morgan, Ward, & Castet, 1998). This suggests that, by 150–200 ms, performance in our spatial coincidence condition is not affected any longer by factors related to the constraints of the dual-task paradigm. 

Figure 3

 

Results of Experiment 1. Orientation discrimination thresholds are plotted as a function of the delay between saccadic cue offset and Gabor patches' onset for the two observers. The relative angle of the tilted patch with respect to the saccadic target is indicated in the legend. Note the different ordinates' scales for the two observers. Error bars have been omitted from the figure for clarity. Asterisks — appearing on this and the subsequent figures — indicate that 95% confidence intervals for the highest and the lowest thresholds of the corresponding curve do not overlap. (a) Results of Experiment 1a (with saccade). (b) Results of Experiment 1b (covert attention).

We next wondered whether the perceptual improvement over time was aligned with the saccadic cue or with the saccade onset. Results for observers EC and SJ are shown, respectively, in the upper and lower parts of Figure 4: Data are replotted as a function of the delay between stimulus and saccade onsets for each condition of relative angle. The highest and lowest thresholds measured in Figure 3a are represented by dashed lines for each condition of relative location in Figure 4. It should be noted that no difference between the thresholds measured in Figures 3a and 4 would be expected if saccadic latencies were similar within each condition. In the extreme case, if latencies were all the same, say 250 ms, then there would be a one-to-one correspondence between delays after the cue and delays before saccade onset (e.g., +50 ms would always correspond to −200 ms). It is only if the distribution of saccadic latencies is broad enough, by chance, that one can hope to show any difference between data plotted in terms of delay after the cue and those plotted in terms of delay before saccade onset. 

Figure 4

 

Reanalysis of data obtained in Experiment 1 for observers EC (top) and SJ (bottom). Orientation thresholds are now plotted as a function of the delay between Gabor patches' onset and saccade onset for each condition of relative angle. In each graph, the upper and lower dashed lines represent, respectively, the highest and lowest thresholds obtained in the previous figure (i.e., when the data were shown as a function of the delay between cue offset and Gabor patches' onset). Error bars correspond to 95% confidence intervals. For both observers, the distribution of saccadic latencies corresponding to each condition of relative angle is shown below the thresholds graphs.

For observer EC (Figure 4), the curves are systematically contained within the two dashed lines indicating that the amplitude of the effect is now smaller than that reported in Figure 3. This flattening of the curve is observed in the three conditions for the long delays before saccade onset. Moreover, the flattening is especially obvious in the +45-deg condition where thresholds are constant in the 150-ms period preceding saccade onset. The difference between conditions seems to be accounted for by the different distributions of saccadic latencies shown below the results for each observer (Figure 4). In the +45-deg condition, latencies are almost uniformly distributed over a wide range so that a given delay before saccade onset, especially a long one, actually includes very different delays after the cue. For instance, a delay of −250 ms could arise from a +50-ms delay with a saccade latency of 300 ms or from a +150-ms delay with a latency of 400 ms, thus resulting in the averaging of performance over the +50- and +150-ms delays. In contrast, in the 0-deg condition, latencies are more narrowly distributed (∼240 ms) so that data plotted as a function either of delay after the cue or of delay before saccade onset tend to become similar in keeping with the smaller flattening observed in this condition. 

For observer SJ, the amplitude of the effect is very close to that obtained in Figure 3 in conditions +45 and −45 deg. This is again consistent with the narrow distribution of latencies (here, ∼210 ms) in these two conditions. With the broader distribution obtained in condition 0 deg, the curve is again flattened. 

In summary, the amplitude of the effect of time on perceptual performance is largest in Figure 3a (i.e., when data are represented as a function of the delay between stimulus onset and cue offset). The same data replotted as a function of the delay from saccade onset (Figure 4) show a smaller amplitude of the effect, especially when the distribution of saccadic latencies is broad. This comparison therefore suggests that it is the cue that triggers attentional deployment. 

The data displayed in Figure 3a also show an overall better orientation acuity when the tilted patch is displayed in the lower visual field rather than in the upper visual field. This spatial inhomogeneity is moreover reflected in Figure 5, which plots, for the three relative locations of the tilted patch, the percentage of trials rejected because saccadic accuracy was not within the tolerated range (within 20% of the required amplitude). For both observers, making a horizontal saccade and concurrently deploying attention in the upper visual field (relative angle = +45 deg) had a detrimental effect on saccadic accuracy. In contrast, saccadic accuracy, when the tilted patch was in the lower visual field (relative angle = −45 deg), was not much affected and was actually similar to the saccadic accuracy obtained when the tilted patch and the saccadic target coincided (relative angle = 0 deg). The poor saccadic accuracy in the +45-deg condition indicates that the poor perceptual performance observed in this condition is not a result of a tradeoff in favor of the saccadic task at the expense of the perceptual task. On the contrary, deploying attention in the upper visual field (while making a horizontal saccade) degrades both tasks. 

Figure 5

 

Percentage of inaccurate saccades in Experiment 1. When the actual saccade was not within 20% of the required horizontal amplitude, the trial was cancelled (and an auditory feedback was provided). For both observers, the percentage of these inaccurate saccades was highest when the tilted patch was in the upper visual field (relative angle = +45 deg). Ninety-five percent confidence intervals are smaller than the size of the symbols.

For the +45- and −45-deg conditions, it was important to verify that observers' saccades did not deviate toward the perceptual targets. We therefore checked offline that the saccades were horizontal as required. For both observers, the median angular deviation from the horizontal was not significantly different from 0 deg. Moreover, the distribution of angular deviations was quite narrow (average interquartile range = 4 deg). This was only slightly larger than that obtained when observers made horizontal 6-deg saccades without any perceptual task to perform (average interquartile range = 2 deg). This high saccadic precision results from the high level of practice in executing accurate saccades for observers SJ and EC. 

Discussion

In the present dual-task paradigm, perceptual performance is best when the saccadic target and the perceptual target coincide, thus confirming previous evidence (Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler et al., 1995). We now show that the enhancement of processing around the saccadic target location gradually builds up over time as reported in studies on covert attention (Cheal & Lyon, 1991a, 1991b; Kinchla, 1992; Murphy & Eriksen, 1987; Nakayama & Mackeben, 1989). 

As already emphasized in a previous study (Kowler et al., 1995), some attention can be diverted from the saccadic target to process stimuli presented at distant locations in the visual field. We capitalized on this ability to test whether the attentional resource attributed to locations away from the saccadic target became more effective with time. We indeed found that orientational acuity for a tilted Gabor patch displayed in the upper or lower visual field (+45 or −45 deg), while a horizontal saccade was in preparation, improved with time. This provides evidence that attention is not locked onto the saccadic goal during the period of saccadic preparation, in contrast to recent suggestions (Deubel & Schneider, 2003). 

The differential processing of stimuli presented either in the upper or in the lower visual field is consistent with previous attentional studies (He, Cavanagh, & Intriligator, 1996; Intriligator & Cavanagh, 2001; Previc, 1990). This attentional signature further confirms that the lower performance for stimuli not coinciding with the saccadic target does result from an attentional factor. Moreover, it shows that future studies on attentional deployment during saccadic programming should not average performance over stimuli presented in different parts of the visual field along the vertical dimension. 

Experiment 1b: Covert attention

The results of Experiment 1a do not allow us to assess whether the overall weak perceptual performance in the ±45-deg conditions is, as we assume, the result of attention being allocated to saccadic programming. It could indeed be argued that this poor performance actually results from visual inhomogeneities in the visual field. It is known that visual sensitivity is poorer when visual stimuli are presented off the horizontal meridian (Carrasco, Talgar, & Cameron, 2001), which is the case when the tilted patch appears at ±45 deg. To control for this confound, we therefore run an experiment that was identical to Experiment 1a except that observers were instructed to keep fixating the center once the cue (pointing either leftward or rightward) was displayed (for one frame). In this classic covert attention task, observers had to shift attention to one of three relative locations (0, +45, or −45 deg: indicated at the beginning of the block) on the side (left or right) validly cued by the central arrow. The absence of saccade was checked after each trial. 

Results and discussion

Orientation discrimination thresholds are shown in Figure 3b for the two observers. The data clearly show that performance at the ±45-deg locations is markedly higher, especially for short delays, than in Experiment 1a, where observers had to make a horizontal saccade. More importantly, the ratio of performance between the ±45-deg locations and the 0-deg location is much smaller in the present experiment than in Experiment 1a. These results therefore suggest that the poor perceptual performance reported in Experiment 1a for locations away from the saccadic goal primarily reflects the cost of preparing a horizontal saccade. These results further suggest that this cost decreases with time as if attentional resources could be gradually withdrawn from the saccadic goal location. Moreover, in contrast with Experiment 1a, the data do not show any significant difference between the +45- and −45-deg conditions (although the latter condition is still producing slightly lower thresholds). This similarity suggests that the large difference favoring the lower visual field over the upper visual field, as reported in Experiment 1a, was actually due to the attentionally demanding selection of the saccadic target. This hypothesis is consistent with previous studies which showed that the upper/lower asymmetry is only reported when the attentional load involved in the task is high (He et al., 1996; Intriligator & Cavanagh, 2001; Previc, 1990). 

Experiment 2: Long duration of saccadic cue

Experiment 1a suggests that the dynamics of attentional deployment is aligned with the saccadic cue (arrow indicating the saccade direction) rather than with saccade onset. In other words, attentional deployment seems to be triggered by the saccadic cue irrespective of saccade latencies. In the present experiment, we aimed at refining this hypothesis and testing whether attentional deployment was triggered by the cue onset or by the cue offset (the latter being the saccadic go signal). If attention is triggered by cue offset, then the temporal course of perceptual performance as a function of the delay after cue offset should be the same irrespective of the cue duration. In contrast, if attention starts its deployment immediately after cue onset, then performance should depend on cue duration: Thresholds measured after a 200-ms cue should be as low as those measured after a delay of 150–200 ms in the previous experiment. 

In the present experiment, the duration of the saccadic cue was therefore set to 200 ms (instead of 6.25 ms as in Experiment 1). The procedure was the same as in the previous experiment except for the two following points:

•  
  The Gabor patches were presented either 6.25 or 150 ms after cue offset (i.e., 206.25 or 350 ms after cue onset).
•  
  The Gabor patch to be attended to was always located above (+45 deg) the saccadic target.

Results and discussion

Results are represented as a function of the delay between the cue onset and the Gabor patches' onset (cue–target onset asynchrony [CTOA]). Gray bars in Figure 6 show the results obtained for the two observers as a function of CTOA. Orientation thresholds are not lowered when CTOA is increased from 206.25 to 350 ms (i.e., when the delay after cue offset increases from 6.25 to 150 ms). These thresholds are slightly smaller than those obtained in Experiment 1 in the last bin of each observer for the same relative location of the tilted Gabor patch (+45 deg)—see the black bars in Figure 6. This very good orientational acuity is consistent with the idea that attention builds up at the +45-deg location during the 200-ms cue duration. The very low percentage of trials that were rejected because of inaccurate saccades (3.8% and 4.6% for EC and SJ, respectively) ensures that this good perceptual performance was not achieved by diminishing the attentional load attributed to saccadic selection. On the contrary, saccadic accuracy is much better here than in the previous experiment for the same condition. 

Figure 6

 

Results of Experiment 2: cue duration = 200 ms. Gray bars show the orientation discrimination thresholds plotted against CTOA (i.e., the delay between the onset of the saccadic cue and the onset of the Gabor patches). Because the saccadic duration is 200 ms in Experiment 2, a CTOA of 206 ms means that the patches appeared immediately after the cue offset. Note, however, that a CTOA of 206 ms in Experiment 1 (black bars) corresponds to the patches appearing 200 ms after cue offset. Error bars correspond to 95% confidence intervals.

Median latencies measured here with the short CTOA were slightly longer than in Experiment 1 for the [0, 50 ms[ bin in the +45-deg condition: 274 versus 255 ms for EC (difference = 19 ms) and 252 versus 211 ms for SJ (difference = 41 ms). 

Could this small increase in latencies be interpreted as evidence that the good perceptual performance is not due to attention building up during the cue presentation but rather depends on a strategy in which observers postpone the programming of saccades? An increase in latencies could suggest that observers were able to reach such a good perceptual performance by postponing the programming of saccades (Kowler et al., 1995). This hypothesis appears unlikely for three reasons. First, it is hard to believe that delaying saccades by 19 ms (EC) or 41 ms (SJ) would be sufficient to halve orientation discrimination thresholds, whereas a 150-ms delay from the cue was necessary in Experiment 1 to achieve the same improvement. Second, if the strategy of delaying saccade by such small amounts was so efficient, one may wonder why observers did not apply this strategy in Experiment 1. 

In summary, the very good performance measured for stimuli displayed right after cue offset (CTOA = 206.25 ms) suggests that attention starts to build up at the perceptual target location as soon as the saccadic cue is displayed. The absence of significant improvement when Gabor patches are displayed 150 ms after cue offset (CTOA = 350 ms) further suggests that attentional deployment saturates at most after 200 ms from the cue onset. In addition, as already suggested by the reanalysis of data in Experiment 1 (Figure 4), these results show that the temporal course of performance in our paradigm is aligned with cue onset rather than with saccade onset. This is moreover consonant with studies on the temporal course of covert attention showing an alignment of data with cue onset (Cheal & Lyon, 1991a, 1991b; Murphy & Eriksen, 1987; Nakayama & Mackeben, 1989). 

Second set of experiments (Experiment 3 and Experiment 4)

The goals of the second set of experiments were (a) to control that the improvement of perceptual performance over time does reflect a spatially specific attentional buildup rather than an alerting effect (Eriksen & Murphy, 1987; Posner, 1980; Remington, 1980) and (b) to verify that temporal masking is a key parameter to reveal this temporal improvement. 

Stimuli

Stimuli were the same as in the first experiment except that the Gabor patches could appear at one of four delays following cue offset: 6.25, 68.75, 131.25, or 193.75 ms. 

The tilted patch always appeared at the saccadic target location except in the spatial uncertainty blocks of Experiment 3, where the tilted patch could appear at any of the eight disks' locations. 

In Experiment 4, the maximal contrast of the white noise dots constituting the masks was 12.5% (instead of 50%). 

Methods

The procedures were the same as in the previous experiments except that adaptive staircase procedures were used to determine the orientation of the tilted patch on each trial (Levitt, 1971). For each of the four postcue delays, two staircases tracked the 0.29 and 0.71 thresholds for clockwise responses. This resulted in eight staircases that were randomly interleaved within a block until 10 valid trials were obtained for each staircase. The collected data were analyzed as in the previous experiments to calculate the orientation discrimination thresholds. 

Observers

The two same observers as in the previous experiments and a third naive observer (NF) participated in the experiments. Observer NF was given approximately 10 hr of practice to minimize perceptual learning over the period of the actual experiments. 

Experiment 3: Effect of spatial uncertainty

In the first set of experiments, it could be argued that, in contrast to our claim, improvement of perceptual performance with time results from a classic nonspecific alerting effect triggered by the cue onset (Eriksen & Murphy, 1987; Posner, 1980; Remington, 1980). It is indeed known that the onset of an endogenous cue (usually in the center of the screen) is able to start a process that improves performance at all locations in the visual field and whose efficiency increases over time (in ∼200 ms). If the temporal improvement we report on was exclusively due to such a warning effect, then presenting the tilted patch at any of the eight disks' locations without precuing its location should still produce the same temporal improvement of performance (i.e., with the same slope). We therefore compared in the present experiment the temporal course of perceptual performance in two conditions: (a) in the first condition, the tilted patch appeared randomly at any of the eight disks' locations (spatial uncertainty); (b) in the second condition, the tilted patch coincided with the saccadic target location on each trial, as in the 0-deg relative-location condition of the first set of experiments. Each condition was run in separate blocks, and observers were informed of the nature of the block they were performing. Our prediction was that the slope of the temporal improvement should be steeper in the latter condition because it should result from the joint influence of the spatially distributed alerting effect and of the rapid spatial focusing of attention. 

Results

In the 0-deg relative-location condition (i.e., when the tilted patch and the saccadic target coincided on each trial), there is a large decrease of thresholds over time for the three observers (filled symbols in Figure 7a) as already observed in Experiment 1: thresholds decreased by as much as 50% for two observers and by 42% for observer EC. For each curve, 1,330 trials on average (including cancelled trials) were collected. However, when the tilted patch randomly appeared at any of the eight disks' locations, there was a much smaller improvement in performance across time for all three observers (large open symbols in Figure 7a). We checked whether this slight improvement was still present when considering only the trials in which the tilted patch was displayed away from the saccadic target. We first excluded from the analysis the trials where the tilted patch had appeared at the saccadic target location (small open symbols with dotted lines). We also excluded trials where the tilted patch had been above (+45 deg), below (−45 deg), or at the saccadic target location (small open symbols with dashed lines). In both cases, the thresholds were similar. These results therefore confirm previous evidence that the cue triggers an alerting effect that develops over time at all locations in the visual field. 

Figure 7

 

Results of Experiment 3: effect of spatial uncertainty of the tilted patch. Solid symbols correspond to the condition where the perceptual target and the saccadic target were in spatial coincidence (observers were informed of this coincidence). Open symbols represent the condition where, on each trial, the perceptual target could appear randomly at one of the eight disks' locations (spatial uncertainty). (a) Orientation thresholds for three observers are plotted as a function of the onset of the Gabor patches (from cue offset). Small open symbols show the thresholds obtained when reanalyzing the data of the spatial uncertainty condition. The dotted lines show the results obtained after excluding from the analysis the trials where the tilted patch and the saccadic target coincided. The dashed lines show the results obtained after excluding from the analysis the trials where the tilted patch was above (+45 deg), below (−45 deg), or at the saccadic target location. Error bars correspond to 95% confidence intervals. (b) Same results normalized with respect to the best threshold of each curve. (c) Normalized thresholds are plotted against log stimulus onset for the three observers. Circles, squares, and triangles correspond, respectively, to observers SJ, EC, and NF.

Thresholds were normalized with respect to the best threshold of each condition to compare the rate of perceptual improvement over time in the two experimental conditions (Figure 7b and c). The clear-cut interaction between the effect of time and the degree of spatial uncertainty of the tilted patch confirms our interpretation. In the spatial uncertainty condition (open symbols), performance improves over time at all locations with a shallow slope as previously reported (Eriksen & Murphy, 1987; Posner, 1980; Remington, 1980). The slope of the improvement is, however, much steeper in the 0-deg relative-location condition (filled symbols), thus revealing an additional effect that we interpret as a rapid deployment of attention toward the cued location. In other words, the difference in slopes between the two curves is an estimate of the spatially selective attentional effect after elimination of the alerting effect. 

Experiment 4: Effect of mask luminance contrast

In the literature on attention, a recurrent problem when trying to compare different studies concerns the absence or presence of temporal masking (Morgan et al., 1998). In our paradigm, forward and backward masking were used to shorten the visible persistence of the Gabor patches and restrict their temporal availability to their nominal duration (Coltheart, 1980b). In other words, we predicted that the temporal improvement reported in the previous experiments would be reduced or eliminated if temporal masking was absent or inefficient. This hypothesis is tested in the present experiment by reducing the contrast of the masks and thus lengthening visible persistence. 

Results

The results of the present experiment are shown in Figure 8 (open symbols) along with the results obtained in Experiment 3 in the 0-deg relative-location condition (filled symbols). Thresholds obtained with the low-contrast mask (12.5%) are globally lower than those obtained previously with the 50% contrast mask. Moreover, and most importantly, the effect of time is now absent or weak. Data normalized with respect to the best threshold of each condition show the predicted interaction (only weak for EC) between time and contrast level (Figure 8b and c). Analyses of saccadic accuracy and saccadic latencies show no significant difference between the two contrast conditions (not shown). 

Figure 8

 

Results of Experiment 4: effect of mask contrast. Solid symbols represent the data obtained in the previous experiment in the 0-deg relative-location condition (contrast of the mask = 50%). Open symbols represent the same condition with a mask contrast of 12.5%. (a) Orientation thresholds for three observers are plotted as a function of the onset of the Gabor patches (from cue offset). Error bars correspond to 95% confidence intervals. (b) Same results normalized with respect to the best threshold of each curve. (c) Normalized thresholds are plotted against log stimulus onset for the three observers.

This interaction between mask contrast and time suggests that the patches' visible persistence with our high-contrast mask is shorter than the duration required before attentional enhancement becomes fully efficient. Consequently, the enhancing effect of attention is lowest shortly after the cue and increases thereafter. In contrast, with the low-contrast mask, Gabor patches displayed right after the cue still persist in the visual system when the attentional buildup has reached an efficient level (after ∼150–200 ms). 

General discussion

We used a dual-task paradigm to investigate the interaction between two closely coupled processes: saccadic programming and spatially selective attention. The specificity of our paradigm is that the two crucial pieces of information needed by observers, namely, the respective locations of the perceptual and saccadic targets, are simultaneously delivered. As soon as the cue appears, and not before, observers can start programming the required saccade and, in parallel, start attending to the location of the perceptual target. We argue that this simultaneity is an essential methodological point to truly investigate the parallel activity of the two processes. 

Our study highlights the dynamics of the spatial deployment of attention during saccadic preparation: Perceptual performance dramatically improves (up to 50%) over a period of 150–200 ms from saccadic cue onset either at the location of the saccadic goal—where overall performance is best—or at a neighboring cued location (±45 deg). The improvement of orientational acuity with time at a cued location away from the saccadic target suggests that attention is not locked on a small area surrounding the saccadic target location during the period preceding saccade execution (an opposite conclusion was recently reached by Deubel & Schneider, 2003, a point that will be discussed in the next section). 

The decrease of orientation discrimination thresholds over time is not solely a result of a nonspecific alerting effect triggered by the cue (Eriksen & Murphy, 1987; Posner, 1974, 1980; Remington, 1980). We controlled for this nonspecific component in Experiment 3 by measuring the temporal course of performance when the tilted patch could randomly appear at any of the eight disks' locations (i.e., without precuing). We found that the slope of this temporal improvement was shallower than that produced when the tilted patch location was precued. Cuing thus allows a spatially selective attentional deployment, which occurs more rapidly than the nonspecific alerting effect. Moreover, the strength of the temporal effect is dependent on the efficiency of the temporal masks used to shorten the visible persistence of the Gabor patches (Experiment 4). This confirms the crucial role of temporal masking in experiments investigating the interaction between attentional and temporal factors (Morgan et al., 1998). 

Our results suggest that the visual system can start deploying attention before the disappearance of the endogenous cue (although observers have the explicit instruction to make the saccade only at the time of cue offset). When the saccadic cue is brief (6.25 ms), orientational acuity monotonically improves over a range of 150–200 ms after cue offset. However, when the saccadic cue duration is much longer (200 ms), orientational acuity is as good as the best performance obtained with the brief duration cue. Therefore, the CTOA (i.e., the delay between saccadic cue onset and Gabor patches' onset) seems to be a better predictor of perceptual performance than the time after cue offset. This is consistent with the idea that the buildup of attention can start as soon as the cue is displayed, which is actually the conclusion reached by classic studies bearing on the temporal course of endogenously triggered covert attention (Cheal & Lyon, 1991a, 1991b; Murphy & Eriksen, 1987; Nakayama & Mackeben, 1989). 

In most cases, orientational acuity is higher when the tilted patch is displayed in the lower visual field rather than in the upper visual field. This spatial anisotropy is consistent with experiments showing an advantage for the lower visual field in tasks that require a high degree of focused attention (He et al., 1996; Intriligator & Cavanagh, 2001; Previc, 1990). To interpret these results, it is classically assumed that parietal regions, heavily involved in attentional processes, receive more projections from the early visual areas located above the calcarine sulcus—that is, those receiving projections from the lower visual field (Maunsell & Newsome, 1987). 

Altogether, our study clearly shows that the temporal course of perceptual performance at and near the saccadic target is influenced by a spatially selective attentional process. Two general interpretations of this result as to the nature of this attentional process have to be considered. The first hypothesis is that an automatic and necessary presaccadic attentional shift deployed for the sake of the primary task (i.e., saccadic programming) is jointly used by the visual system to perform the perceptual task (i.e., the secondary task; Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler et al., 1995). It could thus be argued that, in our experiments, attention is widely distributed around the saccadic target (thereby encompassing our ±45-deg locations) as a result of this automatic shift of attention. This large attentional distribution would be able both to allow accurate saccadic programming toward the saccadic target (i.e., the distribution center) and improve perceptual performance at the saccadic target more than at the ±45-deg locations. A second general hypothesis is that the attentional deployment revealed by the perceptual task in our study is independent of saccadic programming. In other words, in the context of dual-task paradigms, any primary task using a cue to direct attention to a specific location (e.g., covert attention) would be susceptible to elicit an attentional deployment that could be used in parallel to perform a secondary task involving some processing at the cued location. In principle, one could test this hypothesis by using a dual-task paradigm where the primary task would not consist any more in making an accurate saccade but rather in performing a perceptual task whose location would be centrally cued. It would thus be possible to assess whether perceptual performance in the secondary task improves over time as it did in our experiments. To our knowledge, this kind of experiment has never been performed. 

In summary, to study the relationship between saccades and attention, we used an efficient dual-task paradigm whose principles have been clearly exposed in previous studies (Deubel & Schneider, 1996; Kowler et al., 1995). We introduced some crucial modifications to assess the temporal course of the attentional deployment during saccadic programming. The stringent control of saccadic accuracy in our study allows us to report with confidence that the very high level of perceptual performance achieved by our observers does not result from poor allocation of attention to the saccadic task. 

Comparison with psychophysical studies

In a recent study, Deubel and Schneider (2003) put forward the strong claim that attention is locked onto the saccadic target as long as the eye movement is not executed. This conclusion was based on the absence of any improvement of perceptual performance as a function of stimulus onset asynchrony (SOA) at cued locations away from the saccadic goal. In their paradigm, SOA was defined as the temporal delay between the onset of the cue, indicating the saccadic target (saccadic cue), and the acoustical go signal allowing observers to make the saccade (the perceptual target always appeared 80 ms after the go signal). An important point in their experiments was that the absolute location of the perceptual target was always cued a long time (1,500 ms) before the saccadic cue. We argue that because of this very long delay, attentional resources could be allocated to the location of the perceptual target well in advance of the saccadic cue. We believe therefore that perceptual performance did not increase with SOA in their work because attention had already been optimally deployed on the perceptual target as the saccadic cue appeared. In contrast, in our paradigm, it is not possible to deploy attention at the perceptual target location before knowing the saccadic target location because both of these locations are simultaneously indicated (by the arrow cue). There is an additional reason that could explain why perceptual performance did not increase with SOA (as defined in their study): the smallest delay between saccadic cue onset and perceptual target onset used in their experiments was actually 180 ms (nominal SOA = 100 + 80 ms after the acoustical signal). However, our results show that attentional resources can be rapidly and efficiently deployed within this 180-ms period. 

The interaction between attentional deployment and saccadic programming has also been investigated in a set of recent studies (Godijn & Theeuwes, 2003; Van der Stigchel & Theeuwes, 2005). The task used in these studies relied on a perceptual comparison between stimuli displayed before and those displayed after a saccade. Consequently, we believe that the attentional processes involved by this particular experimental paradigm are not the same as those involved in our own paradigm. In these studies, performance obviously relied on a form of visual short-term memory, which is known to be controlled by attention (Coltheart, 1980a; Pashler, 1998; Reeves & Sperling, 1986). Notably, the role of attention is clear in the case of trans-saccadic memory (i.e., when observers have to compare visual information displayed before and after a saccade; Irwin & Gordon, 1998). In contrast, no trans-saccadic comparison was required in our experiments. This important difference should be borne in mind when comparing these earlier studies with ours. 

Models and physiology

Low-level attentional modulations

It is now well established that attention can change neuronal activity in several low-level visual cortical areas, including areas V1, V2, V4, and MT/MST (reviewed in Kastner, 2004; Maunsell, 2004). Moreover, there are evidence that response enhancement in extrastriate cortex due to spatially directed attention occurs as early as 70–130 ms after stimulus onset (Heinze et al., 1994; Hillyard, Teder-Salejarvi, & Munte, 1998; Martinez et al., 1999), a time window consistent with our results. Several computational models, based on psychophysical and physiological results, have proposed different implementations of how attention improves processing in low-level areas (Itti & Koch, 2001). One of these models is especially relevant because it is mainly based on psychophysical results obtained in experimental conditions very similar to ours (Lee, Itti, Koch, & Braun, 1999; Lee et al., 1997). First, these authors used a dual-task paradigm: Thresholds were measured for stimuli presented in the periphery while observers performed, or not, an attentionally demanding task at the center of the display. Thus, peripheral thresholds were obtained with very poor attention or with fully focused attention. Second, orientational acuity was one of the low-level tasks that were assessed in these experiments. It is therefore very interesting to note that the orientation thresholds measured in their study and in the present one lie within the same range. For instance, the best thresholds measured by Lee et al. (1997) in conditions of full attention—with high contrast at 4 deg of eccentricity (their Figure 3)—were approximately 1–2 deg. Strikingly, the same absolute values can be obtained in our experimental conditions, which are much less optimal, thus revealing the remarkable efficiency of attentional deployment in our paradigm. More precisely, values of 1–2 deg are often found when the delay after the cue is approximately 150–200 ms, thus suggesting that perceptual performance after approximately 150–200 ms is not affected any longer by the concurrent attentional resource devoted to the selection of the saccadic target. 

Altogether, these similarities lead us to believe that the model proposed by Lee et al. (1997, 1999) is well suited to account for the low-level attentional modulations observed in our work. Although this model is concerned with the processing of luminance contrast, spatial frequency, and orientation, we concentrate here on the mechanism responsible for the attentional improvement of orientational acuity. This improvement might result either from an increased selectivity (i.e., narrower bandwidth) of single neurons or from an increased response gain leaving the bandwidth unchanged. Although this issue is still controversial in the physiological literature (Maunsell, 2004; McAdams & Maunsell, 1999; Spitzer, Desimone, & Moran, 1988), the psychophysical results of Lee et al. (1997, 1999), along with their computational model, support the former hypothesis. The essence of their model is that attention activates a winner-take-all competition among overlapping visual filters (i.e., at a given location), thereby sharpening orientation tuning in cortical areas that mediate basic pattern vision (presumably V1 and V2). 

Attentional control processes

There is now ample evidence that attention can be distributed to multiple locations in the visual field at a given instant (Gobell, Tseng, & Sperling, 2004). How spatial distribution of attention evolves over time is an unresolved central issue that has received growing attention (Bisley & Goldberg, 2003; Ghose & Maunsell, 2002; LaBerge, 2002; LaBerge, Carlson, Williams, & Bunney, 1997). For instance, a recent study conducted in the lateral intraparietal (LIP) area of behaving monkeys suggested that the spatial and temporal dynamics of a monkey's attention rely on the ensemble activity in the LIP area across the entire visual field (Bisley & Goldberg, 2003). It is not the absolute value of a neuron's activity that determines perceptual performance at a given location but rather the relative activity of this neuron with respect to other neurons having different receptive field locations. In this study, LIP neurons gradually responded with different dynamics dictated by factors influencing the spatial deployment of attention. It is usually assumed that the LIP area represents an attentional saliency map that would be able to drive attentional modulation in lower visual areas (Colby, Duhamel, & Goldberg, 1996; Gottlieb, Kusunoki, & Goldberg, 1998). In addition, there is evidence that one important contribution of the LIP area to oculomotor behavior is the selection of targets for saccades (Wardak, Olivier, & Duhamel, 2002). When applied to our dual-task paradigm, these results suggest that the neural processes underlying the control of low-level modulations could reside in the LIP area with one peak concentrated on the saccadic target and a second peak focused on the tilted patch location. 

Acknowledgment

This work was supported by a grant from the French Ministère de la Recherche (ACI-2001). We thank Marisa Carrasco for her helpful comments. 

Commercial relationships: none. 

Corresponding author: Eric Castet. 

Email: [e-mail]. 

Address: 31 chemin Joseph Aiguier, 13402 Marseille cedex 20, France.