Parallel programming model („race model‟)

A more recent account of antisaccade performance is the „parallel programming model‟, or „race model‟ as it is often referred to (Massen, 2004; Munoz & Everling, 2004; Reuter & Kathmann, 2004). Parallel programming accounts of antisaccade performance share some similarities with the LATER model of saccade generation (Carpenter, 1981). As mentioned earlier, the LATER model assumes that at target onset levels of activity in saccade generating neurons begin to rise at a uniform rate until the threshold for triggering a saccade is reached. Drawing on accumulator models of saccade generation (see Carpenter 1981; Carpenter & Williams, 1995; Hanes & Schall, 1996) the parallel processing model assumes that at target onset, a race ensues between neural activity in the prosaccade (exogenous) pathway and in the antisaccade

(endogenous) pathway, with the winner reaching the threshold appropriate for triggering the saccade first. In other words, if activity in the exogenous pathway reaches threshold first, then an erroneous prosaccade towards the target will be made first, but if activity in the endogenous pathway reaches threshold first, then a correct antisaccade will be made to the opposite side. Therefore, the faster a correct antisaccade can be

programmed, the more likely it is to win the race, and be initiated before the incorrect prosaccade towards the target. Importantly, an incorrect prosaccade is often closely followed by a corrective antisaccade but if an antisaccade is made first, a prosaccade would not follow. This is because in correct trials, activity supporting the antisaccade reaches threshold first, thus “winning” the competition and the build up of activity supporting the erroneous prosaccade towards the target ceases. Figure D illustrates what happens during the time course of a correct antisaccade trial and an incorrect

antisaccade trial respectively and shows the neural representation of the race model for a correct and incorrect antisaccade.

Figure D. The time-course for a correct (top left) and incorrect antisaccade (top right),

and the neural representation of the race model for a correct (bottom left) and incorrect antisaccade (bottom right).

Note: in the bottom left diagram, the endogenous pathway supports an antisaccade to the opposite location of the target, and the exogenous pathway supports a prosaccade to the target. In the bottom right diagram, the endogenous program reaches the threshold for saccade triggering, shortly after the error exogenous program gets there, suggesting a corrective antisaccade was made.

According to race model accounts, there is a close relationship between correct antisaccade latency and antisaccade error rate. Any manipulation that differentially affects pro and antisaccade latencies, will ultimately impact on the probability of an antisaccade error being made, as the manipulation will influence the likelihood of one or other of these processes reaching the threshold for saccade triggering first. Conversely, a manipulation that affects prosaccade and antisaccade activity to the same degree, should not impact on antisaccade error rate, as the relative likelihood of either

prosaccade or antisaccade neural activity reaching threshold first remains unchanged. In a series of experiments, Massen (2004) tested the predictions of the „race model.‟ In her first experiment, she compared performance in separate blocks of pro or antisaccades with performance on a mixed pro/antisaccade paradigm in which participants were unable to predict the upcoming saccade on the basis of the previous saccade. The reasoning was that in the mixed block, participants would not be able to predict the upcoming antisaccade (because each trial could be either a prosaccade or antisaccade), so therefore they should take longer to make correct antisaccades, and as a result, antisaccade errors should increase (as the activity in the prosaccade pathway would be more likely to reach the threshold for saccade triggering first). As predicted, correct antisaccade latencies were increased in the mixed paradigm compared to antisaccade latencies in the separate block of antisaccades, as was antisaccade error rate.

In her second experiment, Massen (2004) attempted to exploit the fact that many individuals show asymmetries in antisaccade performance, making more errors when the target appears on one side compared to the other. Based on the findings of Fischer & Weber (1997), Massen predicted that participants‟ prosaccade latencies would be faster to the side that they made most errors to, or that correct antisaccade latencies would be slower to the other side. In other words, if a participant made more antisaccade errors to the left side, then their prosaccade latencies should be faster to the left side (compared to the right) or antisaccade latencies are slower to the right side (or both). A second prediction was that prosaccade latencies should be faster or antisaccade latencies should be slower (or both) to targets positioned at 12° compared to 6°. The results showed that on average participants made more antisaccade errors to the left than to the right; however, there was no difference in errors between the different eccentricities. In

addition, there was no difference in correct antisaccade latencies when comparing left to right, but antisaccade latencies were faster when the target was presented at 12°.

The second experiment designed to exploit asymmetries in antisaccade performance showed only moderate support for parallel processing predictions, as participants with side asymmetries in antisaccade error rate showed shorter prosaccade latencies to the side where many antisaccade errors were made and slower antisaccade latencies to the opposite side.Although varying stimulus eccentricity did not impact on antisaccade error rate, the effect of this manipulation was consistent with parallel programming predictions. As mentioned above, parallel programming models predict that a manipulation which affects prosaccade and antisaccade latencies to the same degree should not result in a change in antisaccade errors because the likelihood of either prosaccade or antisaccade neural activity reaching the threshold for triggering a saccade first remains unchanged. The results of experiment 2 are in line with this prediction, as prosaccade and antisaccade latencies were both reduced by an increase in stimulus eccentricity.

In the final experiment, Massen measured the effect of inhibition of return (IOR) on antisaccade performance. As mentioned in section 3, IOR is defined as reduced attentional priority for information in a region that has recently experienced a higher priority. Its effect is to bias attentional orienting away from previously inspected locations.Massen used exogenous and endogenous cues to see if IOR would impact on antisaccade error rate. In the first part of the experiment, participants performed separate blocks of prosaccades and antisaccades After fixating a central cross, a cue (white asterix) was presented for 300msec at either the location of the upcoming target (cued trials) the opposite location or at the fixation location (neutral). The cue was followed by a gap of 200msec. In order to draw attention back to the central fixation point, the cue reappeared in the centre, replacing the central fixation cross for another 300msec. Then the central fixation cross reappeared for 200msec, and the target stimulus (green circle) was presented either in the centre of the screen, or in one of the flanker boxes to the left or right of centre. In the second part of the final experiment, centrally presented arrows (endogenous cues) were used to cue attention rather than the asterix.

For part one of the experiment, based on the findings of Rafal, Egly, & Rhodes (1994), Massen predicted that if exogenous cues are able to induce IOR, then detection of the stimulus will be slowed. This means that prosaccade and antisaccade latencies will be affected to a similar degree, thus antisaccade errors should be similar in the trials with uncued and cued stimulus presentation. For the second part, Massen reasoned that if endogenous cues (central arrows) induce IOR, then prosaccade latencies should be

slower when trials are correctly cued compared to when the cue is opposite to the target stimulus location whereas antisaccade latencies should be about the same. This would mean a reduction in antisaccade error rate because the exogenous component in the antisaccade task is selectively slowed allowing activity in the antisaccade pathway to reach the threshold for saccade triggering first in the cued condition, compared to in the condition where the cue does not match the target stimulus location. Essentially, activity in the exogenous (prosaccade) pathway is slowed in the condition with cued stimulus, which should lead to reduced antisaccade error rate in this condition.

The results for part one showed that prosaccade and correct antisaccade latencies were slower when the cue was presented at the same location as the target, compared to when the opposite side of the target was cued. There was no effect of cue condition on antisaccade error rate; therefore errors were equivalent between the cue condition and the uncued condition. As predicted, because both pro and antisaccade latencies were affected equally, this resulted in no change to antisaccade error rate. This finding fits into competitive race model predictions.

In part two, prosaccade latencies were longer in the cued condition, compared to when the opposite side of the target was cued, but this effect was considerably smaller for antisaccades. Therefore, antisaccade errors should be reduced in the cued condition, because prosaccade and antisaccade latencies have been affected to a different degree. Specifically, the exogenous component has been slowed, increasing the likelihood that activity in the endogenous pathway (antisaccade) will reach the threshold for triggering a saccade first. Further analysis showed that as expected, antisaccade errors were reduced in the cued condition.

A recent study by Reuter, Herzog, & Kathmann (2006) provides further support for the race models of antisaccade performance. They found that both schizophrenic patients and healthy participants made more errors and were slower to make

antisaccades when a cue had been presented to the side opposite to the target, compared to when no cue was given. This finding conforms to the predictions of race model accounts of antisaccade performance and emphasises the close relationship between speed and accuracy that these accounts have put forward. These findings support race model predictions, as these accounts suggest that an increase in antisaccade latencies will mean an increase in antisaccade errors. However, it is difficult to appropriately test the predictions of the race model without the same manipulation being used on

The results of Massen‟s study are important, as they show that manipulations resulting in increased correct antisaccade latencies (exp. 1) also result in increased antisaccade errors, confirming race model predictions that if the exogenous prosaccade is slowed, there is a greater probability of the endogenously triggered antisaccade winning the competition and reaching the threshold for triggering a saccade first. These findings also highlight that manipulations that affect prosaccade and antisaccade

latencies to the same degree, will not impact on antisaccade errors (exp. 2), as both pathways have been influenced, which has no bearing on the race to saccade triggering.