Dissociable effects of feature expectation on saccades and presaccadic perception

Article activity feed

1. 

   eLife

   Nov 10, 2026

   eLife Assessment

   This valuable study advances our understanding of how feature expectations influence visual perception and eye movements independently of presaccadic attention. The primary findings - showing that feature expectation and presaccadic attention modulate visual sensitivity through distinct, non-interacting processes - are supported by solid evidence from a carefully controlled factorial design and two psychophysical experiments. However, support for the proposed mechanistic interpretation is incomplete, as the experimental design does not fully distinguish expectation effects from stimulus repetition or separate expectation - related changes in eye-movement timing from stimulus-induced inhibition, and certain aspects of the statistical analysis and perceptual threshold procedures require clarification.

   Read the original source
2. 

   eLife

   Nov 10, 2026

   Reviewer #1 (Public review):

   Summary:

   Bortoluzzi and Rohenkohl investigated the interaction between presaccadic attention (PSA) and feature expectation. PSA refers to enhanced perceptual sensitivity or neural responses at the location of an upcoming saccade, and has been taken as evidence for (partial) coupling between spatial attention and the oculomotor system. Although its spatial component has been extensively studied, much less is known about how PSA interacts with feature processing, such as orientation or spatial frequency. This question is particularly interesting as some have proposed that PSA may contribute to visual stability across saccades.

   In two psychophysics experiments, participants prepared a saccade to a cued location while a low-contrast grating, clockwise or counterclockwise, appeared either at the saccade target or …

   Reviewer #1 (Public review):

   Summary:

   Bortoluzzi and Rohenkohl investigated the interaction between presaccadic attention (PSA) and feature expectation. PSA refers to enhanced perceptual sensitivity or neural responses at the location of an upcoming saccade, and has been taken as evidence for (partial) coupling between spatial attention and the oculomotor system. Although its spatial component has been extensively studied, much less is known about how PSA interacts with feature processing, such as orientation or spatial frequency. This question is particularly interesting as some have proposed that PSA may contribute to visual stability across saccades.

   In two psychophysics experiments, participants prepared a saccade to a cued location while a low-contrast grating, clockwise or counterclockwise, appeared either at the saccade target or the opposite location. One orientation was frequent (64%) and the other infrequent (16%), with 20% target-absent (catch) trials. Participants reported target presence and, when present, orientation. The authors found that sensitivity (d') was higher at the saccade target location, showing the well-known presaccadic enhancement. Surprisingly, sensitivity was lower for the high-probability orientation. Importantly, these two effects did not interact. Moreover, presaccadic enhancement increased with time, approaching saccade onset, whereas the low-probability advantage weakened. They also reported that saccade latency was prolonged for the low-probability orientation, an effect reproduced in a follow-up experiment. The authors conclude that PSA enhances perception independent of feature expectation, and additionally, feature expectation influences both perceptual sensitivity and oculomotor responses.

   Strengths:

   (1) The topic addressed here is important for understanding the nature of presaccadic attention (PSA), particularly how PSA interacts with feature processing. The factorial design allows a clear dissociation of the effects of PSA and feature expectation.

   (2) The dissociation between PSA and feature expectation was examined in multiple ways. In addition to testing their overall interaction, the analysis of their temporal profiles provides further evidence that the two effects may reflect distinct processes. Several statistical analyses were also conducted to rule out differences in saccade latency as an explanation for the lack of interaction.

   (3) The investigation of stimulus probability on saccade latency provides an interesting test of how feature information may influence oculomotor responses.

   Weaknesses:

   (1) The manipulation of feature expectation involves both goal-driven and stimulus-driven components. On one hand, participants were explicitly informed that one orientation was more probable throughout the experiment, which could potentially induce feature-based attention toward the more probable orientation. On the other hand, this manipulation necessarily resulted in greater exposure to one orientation than the other. This difference in exposure could potentially account for the lower sensitivity observed for the high-probability stimulus, through repetition suppression or longer-term sensory adaptation. If the feature-probability effect partly reflects such stimulus-driven mechanisms, its independence from spatially driven presaccadic attention may be less surprising. This interpretation warrants more discussion. More broadly, it remains unclear whether feature expectation would still be independent of PSA if expectation were manipulated in a way that better isolates the top-down aspect of the manipulation, or by manipulating feature-based attention cues.

   (2) The effect of feature probability on saccade latency may be considered in relation to the broader literature on oculomotor interference. Previous studies have shown that visual events during saccade preparation can delay saccade initiation and even alter saccade trajectories. Although the present manipulation differs from classic findings, the interference (by the low-probability stimulus), if any, is presented at the target location. Even so, this literature may provide a framework for interpreting the latency effect. It would also be informative to examine whether feature probability affects saccade trajectories.

   Importantly, the fact that stimulus probability affected saccade onset timing leads to the question regarding whether stimulus probability also affect manual reaction time, the interval between stimulus onset and the button press indicating the perceptual decision. It is important to rule out a potential speed-accuracy trade-off before concluding that stimulus probability affects perceptual sensitivity.

   (3) The descriptions of the implementation of the 2-ADC model are very sparse, and largely delegate it to the reference of the original papers Sridharan et al., 2014 and 2017. I suggest that, to serve as a stand-alone paper, there should be more detailed descriptions about how the model was implemented, and how the d' and criterion were calculated from the full contingency table. I also have reservations, or uncertainty, regarding how the current experimental design can be directly fitted into Sridharan et al.'s original 2-ADC framework. In the original Sridharan et al. model, when the target is present, they consider a design where the "alternative" involves only one dimension with two mutually exclusive alternatives, which is target present at location 1 versus target present at location 2. This allows them to model the internal responses in a two-dimensional space, one dimension for one location (with possible extension to N locations/channels).

   In the present study, the authors instead used the two orientation categories as the two ADC alternatives, which is in principle compatible with the feature-based extension discussed by Sridharan et al. However, the experiment is more complex, adding an additional dimension: the spatial location. Either orientation can occur at either of two locations, with target occurrences at the two locations being mutually exclusive. Thus, the complete stimulus space contains four signal states (High vs. Low probability orientation X 2 locations) plus target absence. That is, the sensory evidence that has to be considered includes four variables: oriH_loc1, oriH_loc2, oriL_loc1, oriL_loc2. It is presently unclear why and how this higher-dimensional problem is decomposed or reduced to a 2-ADC framework by assuming only the two orientations matter, while the location seems to be ignored. The authors should clarify the mapping from stimulus and response contingencies onto the 2-ADC model and justify their use of this framework.

   One could potentially treat the two locations as two independent tasks and apply the 2-ADC framework separately to each location. However, this would require the two locations to be statistically independent, including allowing targets to be present at both locations. In the current design, the locations are coupled: target-absent trials require both locations to be empty, and the presence of a target at one location implies its absence at the other. Conversely, evidence that one location is target-absent increases the probability that the target is present at the other location (given the contingency described in Figure 1b). It would therefore be helpful for the authors to clarify the rationale for this design and justify the appropriateness of applying the 2-ADC model for estimating d' and criterion.

   Read the original source
3. 

   eLife

   Nov 10, 2026

   Reviewer #2 (Public review):

   Summary:

   This manuscript investigates a timely question concerning how expectations about visual features relate to changes in perceptual sensitivity that occur before an eye movement. The authors manipulated the probability of two target orientations while participants prepared a visually guided eye movement and measured both perceptual sensitivity and eye-movement behavior. They report enhanced sensitivity both for targets presented at the upcoming eye-movement goal and for low-probability orientations, with no interaction between these effects. Low-probability orientations additionally delayed eye-movement initiation when they appeared at the movement goal, but not when they appeared at the opposite location. A follow-up experiment reproduced the latency difference for targets presented at the movement …

   Reviewer #2 (Public review):

   Summary:

   This manuscript investigates a timely question concerning how expectations about visual features relate to changes in perceptual sensitivity that occur before an eye movement. The authors manipulated the probability of two target orientations while participants prepared a visually guided eye movement and measured both perceptual sensitivity and eye-movement behavior. They report enhanced sensitivity both for targets presented at the upcoming eye-movement goal and for low-probability orientations, with no interaction between these effects. Low-probability orientations additionally delayed eye-movement initiation when they appeared at the movement goal, but not when they appeared at the opposite location. A follow-up experiment reproduced the latency difference for targets presented at the movement goal.

   The combination of perceptual and eye-movement measures is a strength of the study, and the reported effects are interesting. However, I think the interpretation of both the perceptual probability effect and, particularly, the eye-movement latency effect requires reconsideration. In its current form, the study provides solid evidence for effects of feature probability on perceptual sensitivity and eye-movement timing, but incomplete evidence that these effects reflect the proposed mechanisms.

   Strengths:

   The manuscript addresses an interesting and timely question by bringing together two areas that have largely been studied separately: expectations about visual features and changes in perceptual sensitivity associated with preparation of an eye movement. The experimental design allows the effects of feature probability and eye-movement direction to be assessed within the same task, and the authors complement their perceptual measurements with several measures of eye-movement behavior.

   The main empirical effects are clear. The study reproduces the well-established enhancement of perceptual sensitivity at an upcoming eye-movement goal and demonstrates an additional effect of feature probability, with greater sensitivity for low-probability orientations. The observation that feature probability also relates to eye-movement latency is potentially particularly interesting. The authors additionally conducted a follow-up experiment specifically targeting this latency effect, and they use complementary frequentist and Bayesian analyses to quantify several of their key results.

   Weaknesses:

   A first concern is the interpretation of the feature-probability effect as an effect of expectation. The high-probability orientation was presented four times as frequently as the low-probability orientation. Feature probability is therefore intrinsically confounded with stimulus repetition and adaptation. Greater sensitivity for low-probability orientations could reflect enhanced processing of unexpected information, as proposed by the authors, but it could also arise from attenuation of sensitivity to the repeatedly presented high-probability orientation. The experiment convincingly demonstrates an effect of feature probability, but does not distinguish these mechanisms. This distinction is important for interpreting the effect specifically in terms of prediction or expectation.

   My main concern regards the interpretation of the eye-movement latency results. The authors interpret the approximately 3-ms additional delay for low-probability features at the eye-movement goal in terms of an unexpected feature interrupting or updating the decision process underlying the prepared movement, which they describe as "oculomotor procrastination." However, a visual transient presented during eye-movement preparation is known to transiently inhibit eye-movement initiation. Stimulus-induced oculomotor inhibition therefore provides a more proximal alternative framework for the observed latency effects and is currently not considered. Indeed, the proposed description of newly arriving visual information briefly postponing movement release appears closely related to this phenomenon.

   Relatedly, the spatial specificity of the probability effect does not necessarily establish that feature expectation selectively modifies the prepared action. Perceptual sensitivity is substantially enhanced at the eye-movement goal. Thus, the absence of a probability-dependent latency effect at the opposite location could potentially result from weaker processing of the stimulus there, whereas a better-processed unexpected stimulus at the eye-movement goal may produce stronger stimulus-induced inhibition. The authors show that the interaction remains after accounting for correct versus incorrect perceptual reports, but response accuracy does not equate to the strength of the underlying sensory representation across locations. This issue is particularly relevant to the authors' subsequent discussion of separate visual mechanisms for perception and action. The accumulator and dual-systems interpretations should, in my view, be reconsidered in light of the oculomotor-inhibition account.

   An important related limitation is that the latency analyses compare trials in which the Gabor appeared at the eye-movement goal with trials in which it appeared at the opposite location. The finding that latencies are longer in the former condition therefore does not itself establish that presentation of the Gabor delayed eye movements. Experiment 1 includes target-absent catch trials, which could provide an informative baseline for determining whether target presentation itself produces a latency cost. Characterizing this effect is important for distinguishing the proposed account from stimulus-induced inhibition.

   The interpretation of the temporal analyses also requires caution. Target presentation occurred at a fixed time after the eye-movement cue, whereas its timing relative to eye-movement onset was determined retrospectively by when the eye movement occurred. Critically, the manuscript itself demonstrates that target processing and feature probability alter eye-movement latency. The stimulus used to measure perceptual sensitivity can therefore alter the event relative to which its "presaccadic" timing is defined. Splitting trials separately within conditions avoids a simple imbalance of trials between temporal bins but does not eliminate this dependency. This is particularly relevant because estimates of the time course of perceptual sensitivity around eye movements are known to depend on the stimuli used to measure them. Consequently, the different temporal profiles reported for feature probability and presaccadic enhancement should be interpreted cautiously.

   Finally, the visual stimulation itself is spatially asymmetric. A Gabor is presented at only one of the two possible locations before bilateral masks appear. Target onset therefore provides information about the location containing the perceptual signal and produces a visual transient at one location but not the other. This design makes it difficult to separate effects associated with presaccadic selection from differences in the visual input occurring at versus away from the eye-movement goal. A spatially balanced design in which visual stimuli appear simultaneously at both locations would provide a cleaner test.

   Overall, the empirical findings are potentially valuable for understanding how learned feature probabilities relate to perception and eye-movement behavior. However, the present design does not uniquely attribute the perceptual effect to expectation rather than repetition-related processes, and the latency findings should be evaluated within the established framework of stimulus-induced oculomotor inhibition before stronger conclusions about distinct mechanisms for perception and action are drawn.

   Read the original source
4. 

   eLife

   Nov 10, 2026

   Reviewer #3 (Public review):

   The manuscript "Dissociable effects of feature expectation on saccades and presaccadic perception" addresses an interesting and timely question. It asks whether expectations about upcoming stimulus features interact with presaccadic perceptual enhancement, or if feature-based expectations and saccade preparation influence perception and action through partly independent routes. The authors use a paradigm where saccade direction and the probability of target orientation are manipulated independently. The core finding is that perceptual sensitivity is higher at the saccade target and that sensitivity is increased for less expected feature orientations, but these two effects do not interact. Unexpected target features at the saccade goal do, however, produce longer saccade latencies, and this effect persists …

   Reviewer #3 (Public review):

   The manuscript "Dissociable effects of feature expectation on saccades and presaccadic perception" addresses an interesting and timely question. It asks whether expectations about upcoming stimulus features interact with presaccadic perceptual enhancement, or if feature-based expectations and saccade preparation influence perception and action through partly independent routes. The authors use a paradigm where saccade direction and the probability of target orientation are manipulated independently. The core finding is that perceptual sensitivity is higher at the saccade target and that sensitivity is increased for less expected feature orientations, but these two effects do not interact. Unexpected target features at the saccade goal do, however, produce longer saccade latencies, and this effect persists after accounting for perceptual reports and is reproduced in a second experiment.

   A big strength here is the design itself. By tweaking saccade direction and feature probability independently, the researchers directly tested whether expecting a feature changes how strong pre-saccadic enhancement is. They are careful about possible confounds, using ANCOVAs to control for saccade latency and endpoint accuracy, indicating that the main perceptual effects are not just due to timing or landing precision. Pre-saccadic enhancement gets stronger as the target nears saccade onset, while the perceptual peak for low-probability features is stronger earlier and fades as saccade onset gets closer.

   Another positive point is the saccade-latency analysis. The finding that low-probability features push up latency when shown at the saccade target is pretty interesting. It hints that feature expectation can influence ongoing motor preparation even after a direction cue has instigated the motor plan. The generalized linear mixed-model approach is helpful, showing the location-specific feature-probability effect holds even after factoring in perceptual reports. The latency difference, roughly 3 ms in the model-adjusted estimates, is small, but it is statistically significant in the main experiment.

   Trying to reproduce this latency effect in a second experiment adds value too. In that simpler setup, with very visible stimuli, no catch trials, and the target always at the saccade goal, latencies still came out longer for low-probability features. That suggests the effect is not wholly tied to the perceptual-report demands of the first study.

   (1) The Experiment 2 bootstrap p-value needs clarification

   The authors resample latency values within each participant and condition, compute the median difference, and then average the participant-level differences. They do this 10,000 times and set the one-tailed p-value to the proportion of bootstrap iterations where the group difference is < 0. This bootstrap distribution is centered on the observed data, not on a true null distribution centered at zero. However, the proportion of bootstrap estimates {less than or equal to} 0 does not necessarily correspond to a conventional hypothesis-test p-value. Please clarify how pboot was calculated and why this approach provides valid frequentist inference for a null hypothesis of zero difference.

   (2) The staircase procedure estimates a conditional detection threshold, not necessarily "50% detection performance"

   The paper states that a 1-up/1-down staircase yielded 50% detection performance, but then says the staircase was updated using only the 48 signal-present trials, based on hits and misses. False alarms were not used to update the staircase.

   Because the experiment includes signal-absent catch trials, this distinction is important. If false alarms are not included, the staircase is estimating the contrast tied to about 50% conditional detection when a target is present, rather than a true 50% threshold in a full signal-detection framework. The authors should clarify what they mean by "50% detection performance" and why this staircase approach is appropriate for calibrating the subsequent 2-ADC analysis.

   This choice could directly affect the main perceptual finding-the sensitivity advantage (d′) for low-probability features. Since target contrast was fixed based on conditional hit rates (P("Present"∣Signal)) rather than a bias-free detection index (d′), the procedure might not guarantee that baseline sensitivity was equalized across conditions before the main task. If participants shifted their decision criterion (C) depending on feature expectation, or if not tracking false alarms biased the threshold estimate, the baseline could artificially inflate or distort the d′ differences between high- and low-probability conditions. The authors should address whether this staircase could be a potential confound for the main perceptual effects.

   (3) Temporal analysis and its limitations

   The timing analysis splits pre-saccadic timing into Early and Late conditions using a participant- and condition-specific median split. While simple, this reduces temporal resolution and can hide nonlinear or continuous relationships between timing and perceptual sensitivity. A continuous analysis of target offset relative to saccade onset, for example using regression or a mixed-effects model, could provide a clearer picture of the temporal dynamics. It might also reveal whether the apparent weakening of the feature-probability effect is gradual or concentrated in a specific pre-movement window.

   (4) Absence of a significant interaction and Bayes factor

   The lack of a significant interaction between feature probability and saccade direction is supported by a Bayes factor of about 3 in favor of the null. That is informative, but it should not be taken as definitive proof that the two mechanisms are completely independent. I would instead describe this as evidence against a substantial interaction under the tested conditions.

   Read the original source
5. 

   eLife

   Nov 10, 2026

   Author response:

   We thank the Editors and reviewers for their careful reading of our manuscript and their constructive comments. We appreciate their positive assessment of our work and recognize the importance of the concerns they raise. In the revised manuscript, we will address the potential contribution of stimulus repetition and adaptation by introducing additional control analyses examining how preceding stimuli and responses influence perceptual reports. These analyses will help assess the extent to which recent stimulus and response history may contribute to our findings. Furthermore, we will provide a more detailed explanation and justification of the two-alternative detection-choice (2-ADC) model, including how sensitivity (d′) and response criteria were estimated from the stimulus–response contingencies. Additional …

   Author response:

   We thank the Editors and reviewers for their careful reading of our manuscript and their constructive comments. We appreciate their positive assessment of our work and recognize the importance of the concerns they raise. In the revised manuscript, we will address the potential contribution of stimulus repetition and adaptation by introducing additional control analyses examining how preceding stimuli and responses influence perceptual reports. These analyses will help assess the extent to which recent stimulus and response history may contribute to our findings. Furthermore, we will provide a more detailed explanation and justification of the two-alternative detection-choice (2-ADC) model, including how sensitivity (d′) and response criteria were estimated from the stimulus–response contingencies. Additional methodological and statistical clarifications will address the staircase procedure and the calculation and interpretation of the reported bootstrap p-value in Experiment 2. The reviews also offer several suggestions for improving the presentation, terminology, and clarity of the manuscript. In particular, we will clarify our use of the terms “feature expectation” and “feature probability,” distinguishing the experimental manipulation from its interpretation in terms of expectation. Finally, we will revise the discussion of the latency results to incorporate the literature on oculomotor interference and stimulus-induced saccadic inhibition highlighted by the reviewers. A detailed response to the reviewers’ comments will accompany the revised manuscript.

   Read the original source
6. 

   Version published to 10.7554/elife.112671.1 on eLife

   Nov 10, 2026
7. 

   Version published to 10.7554/elife.112671 on eLife

   Nov 10, 2026
8. 

   Version published to 10.64898/2026.07.05.735520v1 on bioRxiv

   Jul 9, 2026