Knox, P.C. (in press) Stimulus predictability and the gap effect on pre-saccadic smooth pursuit NeuroReport
Note: figures unavailable in this version.

Stimulus predictability and the gap effect on pre-saccadic smooth pursuit

Paul C. Knox

Vision Sciences,
Glasgow Caledonian University,
City Campus,
Cowcaddens Road,
Glasgow G4 0BA

Tel: 0141 331 3695
Fax: 0141 331 3387


There is disagreement in the literature as to whether smooth pursuit latency is reduced when a temporal gap is introduced between the extinction of a central fixation target and the illumination of an eccentric moving target. This study confirms that in human subjectssmooth pursuit latency is reduced by gaps and that the magnitude of the reduction is related to the duration of the gap. However, latency is not solely determined either by visual factors or by task parameters such as spatial predictability, but is affected by task context. The results suggest a role for non-visual factors such as attention in the initiation of pursuit.

Key Words: smooth pursuit, express smooth pursuit, eye movement, oculomotor control


It has been known for some time that when a temporal gap is introduced between a central fixation target being extinguished and an eccentric target being illuminated, towards which a subject has been instructed to make a saccade, saccade latency is reduced. More recently, there has been disagreement in the literature as to whether smooth pursuit (SP) latency is reduced when there is a gap between a central visual fixation target being extinguished and a moving target being illuminated1,2,3. The issue is important as confirmation of a gap effect on SP latency would imply important similarities in the initiation of saccades and SP and perhaps that attentional mechanisms are involved in both.

Methodological differences might account for some of disagreement in the literature. The gap effect on SP latency has been investigated using a step-ramp task, with and without a temporal gap between fixation target extinction and pursuit target appearance to either the left or right of fixation. This elicits a response usually consisting of SP in the direction of target motion and one or more saccades in the direction of the target step. The first eye movement in response to target appearance might be a saccade or a short episode of SP interrupted by a saccade before SP continues. Some authors have used desaccaded records for their analysis2,3 while others took no steps to deal with the occurrence of saccades1. This might account for the difference in SP latency distributions reported1,2.

The spatial predictability of the standard task (a step randomly to the left or right, followed by motion in the opposite direction back through the centre of the display) is high: as soon as the subject has identified the direction of the target step, they know (potentially) the direction in which the target will move. As both anticipation6 and prediction7 are known to play a role in SP, it is necessary to establish whether stimulus spatial predictability could account for part or all of the reduction in SP latency observed in gap pursuit tasks1,2. If the reduction in SP latency in gap tasks were related systematially to prediction or anticipation then differences in analysis criteria might account for the different results reported.

To address these issues new experiments have been conducted in which the target step amplitude and velocity used ensured a high yield of responses which began with smooth eye movement as opposed to a saccade; the latency of this pre-saccadic SP was analysed. In addition to using the standard task, consisting of runs of centripetal trials as previously1,2, SP latency was examined in runs consisting of a single trial type (where spatial predictibility was maximal) and of mixed centrifugal and centripetal trials (where spatial predictability was minimal).

Materials and Methods

Experiments were performed on three right-handed naive subjects (all male; KSB, aged 26; PM, aged 30; MZ, aged 45) who sat 57cm from a visual display which they viewed with their left eye; the right eye was occluded. Stimuli (generated by a Cambridge Research Systems Visual Stimulus Generator) consisted of a dark fixation square (0.3(x 0.3(; 3cd/m2 on a 45.4cd/m2 background) which was presented for a variable period of between 0.5s and 1.5s, in the centre of the display. This was replaced by the pursuit target which subjects were instructed to follow; the pursuit target was also a dark square with the same parameters as the fixation target. Subjects were presented with runs of 96 trials. In interleaved centripetal runs (CP; the condition used previously1) the target appeared randomly 5( to the right or left of the fixation target position, and moved back through the centre of the display at 14(/sec. In each run, four conditions were interleaved from eight combinations of step/motion direction and gap duration and presented in pseudorandom order. In each set of four there was one task with no gap (the "normal" condition) and three with gaps of 100ms, 200ms or 400ms between fixation target extinction and pursuit target appearance. In two conditions the target moved right, and in two it moved left. Subjects PM and KSB were exposed to two further types of trial run. In mixed centripetal/centrifugal runs (CP+CF), the gap duration was fixed at either 0ms or 200ms and after a 5( step (either to the left or right) the target either moved back through the centre of the display or continued moving in the same direction as the step (centrifugal trials). The four conditions (left step with leftward motion, left step with rightward motion, right step with rightward motion and right step with leftward motion) were interleaved. Single task runs (S) consisted of 96 trials with either no gap or a gap duration of 200ms, a target step to the right, and target motion to the left. Thus data was collected for this particular task (target step to the right, target motion to the left - the probe task), with either no gap or a 200ms gap, "embedded" in three different contexts: single task (S), cetripetal trials only (CP) and combinations of centripetal and centrifugal trials (CP+CF). The level of stimulus spatial predictability varied from S where it was highest, to the CP+CF condition where it was lowest. Horizontal eye movement was recorded by means of an infra-red corneal reflection device (IRIS: Skalar), and the eye position signal digitised with 12-bit precision at 1kHz using a CED (1401. The eye position and a time marker of the appearance of the pursuit target were displayed on the computer screen; data from 100ms before to 500ms after the appearance of the target was stored on disc for analysis off-line. SP latency was measured using an analysis program which displayed the recorded eye position and the time at which the pursuit target appeared. The latency of pursuit was taken to be the time from the appearance of the target to the first smooth eye movement in the appropriate direction. Only trials in which target appearance was preceded by steady fixation were analysed.


All of the results reported here are for the latency of smooth pursuit (SP) from trials in which the first eye movement was a smooth eye movement and not a saccade i.e. for presaccadic SP. The mean latency of SP in the normal task (gap=0ms) was similar for all three subjects at around 200ms (eg for SP to the left, KSB:200(27ms, PM:192(26ms, MZ:209(34ms; mean(s.d; Fig. 1). For two of the subjects the mean latency in the normal condition for SP to the left and right was almost identical. For the third subject (MZ) latency was statistically significantly lower for pursuit to the right (195(32ms; "d" test, p<0.001).

Figure 1 near here

The introduction of the shortest temporal gap used in these experiments, 100ms, caused statistically significant decreases in the mean latency of pre-saccadic SP in all three subjects for both leftward and rightward SP. Indeed the effect was remarkably consistent between subjects. The largest reduction observed with this gap duration was 16% (KSB, leftward) and the smallest 13% . Increasing the gap duration slightly reduced SP latency further. While for all but one direction of SP in one subject, there were small increases with gaps of 200ms, in most cases it decreased with the longest gap duration used, 400ms. For all three subjects the lowest mean latencies were observed with the longest gap (eg for SP to the left, KSB:166(30ms, PM:166(29ms, MZ:158(29ms). The lowest mean latency of all was 136(34ms (PM rightward) a reduction of 29% from the normal condition. Correlation coefficients were all statistically significant (p<0.001) for the relationship between gap duration and SP latency for both directions in all three subjects.

For one subject (KSB) there was no significant difference at any gap duration between the latencies for leftward and rightward SP, while for another subject (MZ) there was a systematically statistically significant (p<0.001) lower latency for rightward SP in all but the longest gap condition. The third subject (PM) was intermediate between these two.

Figure 2 near here

An examination of the distribution of SP latency showed no tendency to divide into two or more modes when gaps were introduced. Figure 2 shows a typical set of latency distribution histograms (subject PM, SP to right). As these also show, there were relatively few trials where there was evidence of anticipation (latency <100ms).

Figure 3 near here

In Figure 3, SP latency for the probe task, with and without a gap of 200ms, is shown for the three task contexts used, for the two subjects tested. As expected, latency was lowest in the S condition. Comparing S with CP (the standard task used previously1 and by other authors2,3), in the normal condition (gap=0ms) there was a 14% reduction in latency for PM and a 20% reduction for KSB. Both reductions were statistically significant (p<<0.001). In the case of PM there was also a statistically significant reduction (p<0.001) in latency when a 200ms gap was introduced in the S condition. The unexpected result was that SP latency for the probe task in the mixed CP+CF condition was lower than for the CP condition both with and without a 200ms gap. For PM, SP latency in the CP+CF condition without a gap was 182(20ms, compared to S:166(22ms and CP:192(26ms. The reductions in latency in the S and CP+CF conditions from the CP condition were both statistically significant (p<0.001 and p<0.01, respectively). For subject PM the latency in all three conditions was further statistically significantly reduced with a 200ms gap (p<0.001). The same general pattern was demonstrated by subject KSB. However, the addition of a gap signifcantly reduced latency in the CP condition only.


The results reported here for pre-saccadic pursuit are broadly in agreement with previously published accounts based on both desaccaded2 and non-desaccaded1 records. The absolute SP latency values are lower in the current experiments than reported previously for similar tasks1, and the reductions in latency from the normal condition when gaps are introduced are slightly smaller. However, the visual conditions and stimulus parameters in the two sets of experiments were different. The higher target contrast and target velocity used in these experiments would be expected to reduce latency8,9. Whether these factors could also explain the smaller gap effect than previously observed is currently being investigated.

A gap effect, that is a reduction in the latency of SP latency from the normal condition, was observed even for the shortest gap duration employed (100ms). This confirms previous reports1,2 while leaving open the question of why one group3 failed to find such an effect. The magnitude of the gap effect is modified by the duration of the gap and slightly increases up to the longest gap duration used (400ms). The lack of multimodality in the latency distributions in gap tasks suggests that the earlier observation of this1 may have been an artifact resulting from the use of non-desaccaded records. However, it should be noted that this negative result does not remove the possibility of the existence of a low-latency class of SP responses. The existence of express saccades was disputed by a number of authors who failed to find multimodal saccade latency distributions10. Several factors might influence the pattern of distribution observed, and while it seems unlikely that there is a phenomenon of ìexpressî SP, the history of the express saccade suggests caution in ruling out the possibility altogether.

If stimulus spatial predictability were a key determinant of the gap effect on SP latency, then latency should have been be low in S (in which subjects learned in a few trials the amplitude and direction of the step and the direction and speed of target motion), intermediate in CP (because once the direction of the step had been identified the direction of motion and the target speed was always the same) and longest in CP+CF where the uncertainty about the target was greatest (identifying the direction of the step did not uniquely identify the direction of target motion). The lowest latencies were indeed observed in S, although there was little evidence of anticipation. Note that there was an element of temporal uncertainty in S as the length of the fixation period varied randomly between 0.5s and 1.5s. If this temporal uncertainty is removed, anticipation may play a more prominent role11. Even in S, the presence of a 200ms gap further reduced latency in one subject and slightly reduced it in the other.

The longest latencies were observed in CP, not in CP+CF where predictability was lowest. One possible explanation is that subjects found the CP+CF condition extremely demanding. Subjects found centrifugal trials very difficult and reported concentrating harder in runs consisting of both centrifugal and centripetal trials. Perhaps this increase in attentional resources which they brought to bear in the CP+CF condition had the net effect of reducing latency in the probe task (which was itself an ìeasierî centripetal task), below that observed in the CP condition. Certainly this result suggests that the SP latency is not determined by the visual conditions or the gap duration alone; other factors, perhaps attentional factors, are involved. p>


These results confirm that there is a gap effect on smooth pursuit latency but no clear phenomenon of "express" smooth pursuit. The gap effect cannot be explained in terms of prediction or anticipation as even when subjects know the position of target appearance, the direction of target motion and its speed, there is relatively little evidence of anticipation and there may still be a gap effect. The spatial predictability of the task does not determine SP latency either. The results suggest that higher level factors such as attention may play a role in SP initiation. However, further experiments are required to examine this issue directly.


1. Knox, P.C. NeuroReport 7, 3027-3030 (1996).

2. Krauzlis, R.J. and Miles F.A. J. Neurophysiology 76, 2822-2833 (1996).

3. Morrow, M.J. and Lamb N.L. Exp. Brain Res 111, 262-270 (1996).

4. Saslow, M.G. J Opt Soc Am 57, 1024-1029 (1967).

5. Merrison, A.F.A. and Carpenter R.H.S. Vision Res 35, 1459-1462 (1995).

6. Kao, G.W. and Morrow M.J. Vision Res 34, 3027-3036 (1994).

7. Barnes, G.R. and Asselman P.T. J. Physiol. 439, 439-461 (1991).

8. Tychsen, L. and Lisberger S.G. J. Neuroshysiol. 56, 953-968 (1986).

9. Carl, J. and Gellman R. Journal of Neurophysiology 57, 1446-1463 (1987).

10. Kingstone, A. and Klein R.M. Percept Psychophys 54, 260-273 (1993).

11. Knox, P.C. J. Physiol 495, 139P (1996).

Figure Legends

Figure 1 Plots of SP latency against gap duration for three subjects. Plotted as mean(s.e.m. Squares and solid lines: SP to left. Circles and dotted lines: SP to right. Asterisks (*) mark those conditions in which the latencies to right and left are statistically significantly different (d test, p<0.01).

Figure 2 Frequency distribution histograms of smooth pursuit (SP) latency for subject PM, plotted as percentage of total number of observations in each condition. Data for SP to right is shown for the four gap durations. Figures on the top right of each histogram are top:mean; middle: SD; bottom: (number of observations).

Figure 3 Mean SP latency ((s.e.m.) for the probe task (target step to the right, target motion to the left) gap duration either 0ms or 200ms, for two subjects (PM and KSB) when presented in runs consisting of different combinations of task types. CP (circles and dotted line): centripetal tasks only; CP+CF (diamonds and dashed lines): mixture of centripetal and centrifugal tasks; Single (squares and solid lines): probe task alone.