Elsevier

Neuroscience

Volume 452, 1 January 2021, Pages 345-353
Neuroscience

Research Article
Modelling Prosaccade Latencies across Multiple Decision-Making Tasks

https://doi.org/10.1016/j.neuroscience.2020.11.022Get rights and content

Highlights

  • LATER (Linear Approach to Threshold with Ergodic Rate) can model reaction times.

  • Different eye movement tasks have used different versions of the LATER model, however.

  • We find components common to each model have common parameters across some tasks.

  • Our findings suggest common machinery underlies certain eye movement decision tasks.

Abstract

Oculomotor decision making can be investigated by a simple step task, where a person decides whether a target has jumped to the left or the right. More complex tasks include the countermanding task (look at the jumped target, except when a subsequent signal instructs you not to) and the Wheeless task (where the jumped target sometimes then quickly jumps to a new location). Different instantiations of the LATER (Linear Approach to Threshold with Ergodic Rate) model have been shown to explain the saccadic latency data arising from these tasks, despite it being almost inconceivable that completely separate decision-making mechanisms exist for each. However, these models have an identical construction with regards to predicting prosaccadic responses (all step task trials, and control trials in countermanding and Wheeless tasks, where no countermanding signal is given or when the target does not make a second jump). We measured saccadic latencies for 23 human observers each performing the three tasks, and modelled prosaccade latencies with LATER to see if model parameters were usefully preserved across tasks. We found no significant difference in reaction times and model parameters between the step and Wheeless tasks (mean 175 and 177 ms, respectively; standard deviation, SD 22 and 24 ms). In contrast, we identified prolonged latencies in the countermanding tasks (236 ms; SD 37 ms) explained by a slower rise and an elevated threshold of the decision making signal, suggesting elevated participant caution. Our findings support the idea that common machinery exists for oculomotor decision-making, which can be flexibly deployed depending upon task demands.

Introduction

A variety of oculomotor tasks exist that differ in the complexity of the decisions they require. One of the simplest is the step task, when a fixation target jumps to either the left or right and the observer decides in which direction to make a saccade to follow it. More complex is the countermanding task, in which participants look towards an abruptly appearing peripheral target except when a subsequent cue – the countermanding signal – tells them not to (Hanes and Schall, 1995, Hanes and Carpenter, 1999, Emeric et al., 2007, Salinas and Standford, 2013). In the Wheeless (or pulse-step, or double step) task, an abruptly appearing target sometimes quickly jumps to a new location, and so participants either first look to the initial location or they override this response and saccade directly to the new location (Wheeless et al., 1966). Both the countermanding and Wheeless tasks involve the need to stop a decision to perform an action that is no longer appropriate. Indeed, the countermanding task can be viewed as a version of the Stop-Signal Paradigm, outlined by Logan and Cowan (1984). Often the need to promptly halt actions is of critical importance in the natural world: for example, halting the decision to depress your car’s accelerator after the traffic signal has turned green, because a pedestrian then steps out in front of you (Noorani, 2017).

Decisions are commonly believed to arise through some form of race between neural processes representing different outcomes, and many models exist to explain response times for such decisions (Ratcliff, 1978, Carpenter, 1981, Ratcliff and Rouder, 1998, Brown and Heathcote, 2008). Commonly, one of the competitors in the race is a stop signal designed to halt actions that are no longer appropriate, such as in Wheeless (Camalier et al., 2007) and countermanding (Band et al., 2003) tasks. Here we concentrate on one particular model – LATER (Linear Approach to Threshold with Ergodic Rate) – which has been successfully used to model a variety of oculomotor decision tasks using relatively few parameters. However, one undesirable consequence is that specific instantiations of the LATER model exist for different decision tasks (Carpenter and Williams, 1995, Asrress and Carpenter, 2001, Carpenter et al., 2009, Noorani and Carpenter, 2015) despite it being almost inconceivable that each task involves the operation of completely separate decision mechanisms.

An example of different instantiations of the LATER model for three oculomotor decision tasks mentioned above can be seen in Fig. 1. In each model, the appearance of the peripheral target causes a decision signal S to rise linearly until a threshold ST is reached, at which time a saccade toward the peripheral target is made. On any given trial, the rate of rise r is selected from a Gaussian distribution of mean μ and standard deviation σ. In LATER, each decision alternative is represented by a similar rise to threshold unit, with the final decision determined by which unit wins the race to threshold (Carpenter, 1981, Carpenter, 1999). In the case of the countermanding and Wheeless tasks, additional LATER decision units are activated in response to either the appearance of a countermanding signal or a jump in the peripheral task location respectively, and the action of these units is to stop the original saccade from taking place and, in the case of the Wheeless task, to trigger an alternative response. The average time for these stopping units to reach threshold is reduced (Noorani and Carpenter, 2015), thereby helping to ensure inappropriate actions are cancelled with a high degree of reliability. However, as each unit has an independent rates of rise r (Leach and Carpenter, 2001), there is the possibility that the LATER unit responsible for making a saccade to the initial target may win the race against the stop unit, thereby resulting in an error in the countermanding task, or a saccade towards the initially appearing target in the Wheeless task. Although it has been suggested error responses in decision experiments could only be accounted for via subsequent embellishments to the LATER model (Brown and Heathcote, 2008, Sun and Landy, 2016), the idea of an independent race between decision alternatives – and, therefore, the possibility of errors – was inherent in the earliest statistical descriptions of the model (Carpenter, 1981, Carpenter and Williams, 1995). Confusion may have arisen given that the LATER acronym was only applied to the model in 1999 (Carpenter, 1999), and that early work involved experimental paradigms where errors were largely absent because one alternative almost always won the race (e.g. simple choices involving high contrast targets appearing in unambiguous locations) (Carpenter and Williams, 1995). Under such circumstances, the other competitors in the race can be ignored and successful modelling thereby achieved with a single LATER unit.

Of note is that all three models contain some identical constituent components. In particular, the component for a prosaccade to the initially appearing target is identical in all cases. Therefore, it might be expected that saccades in the step task have the same characteristics as control trials in the other two tasks (i.e. non-countermanded trials in the countermanding task, and trials with only a single step in the Wheeless task). However, observers might adopt a higher, more conservative threshold ST (i.e. show decreased urgency (Reddi and Carpenter, 2000)) in the countermanding and Wheeless tasks to delay their initial saccade, and so allow more time for either successful countermanding or to saccade directly to an altered target location. It might be thought that lateral inhibition between different rise-to-threshold units in the models could act to decrease latency by decreasing the average rate of rise μ of the prosaccade decision signal. Lateral inhibition effects have been previously incorporated into LATER (Hanes and Carpenter, 1999, Leach and Carpenter, 2001, Noorani and Carpenter, 2015), and occur when two or more units are activated in the decision process (e.g. the “Go” and “B” units in the Wheeless task; Fig. 1). However, in control trials where only a single unit is activated, substantial lateral inhibition effects would not be expected. Although changes in ST or μ can produce delayed latencies, each produces a characteristic alteration to latency distributions (Carpenter and Williams, 1995), thereby allowing their effects to be separated.

Experimentally verifying that such elements are common (i.e. have the same predicted parameters within a model) is challenging based on the existing literature, as each model was developed from data from different participants whose individual model parameters will presumably differ. Furthermore, stimulus parameters often differ between previously published experiments. It would therefore be of considerable advantage to have a dataset from a group of observers who perform all three tasks, using a common stimulus, so that model parameters between tasks could be better compared. Another potential complication is that modelling work with comparatively small numbers of highly-trained observers (for example, Carpenter and Williams, 1995, Hanes and Carpenter, 1999, Carpenter et al., 2009, and Noorani and Carpenter (2015)) may not give an accurate indication of the variability in responses – and, therefore, model parameters – contained within a larger, less well-trained group.

Here we performed a collection of common eye-movement decision tasks – both simple (step task) and more complex (countermanding and Wheeless) – on a single group of observers. We then modelled saccades from the step task and control trials from the countermanding and Wheeless tasks to investigate whether model parameters in LATER were usefully preserved across tasks. Should such parameters be preserved, this would provide behavioural evidence for the existence of common components underlying decision-making that can then be flexibly deployed depending upon the specific requirements of the task.

Section snippets

Participants

We collected data from 24 participants (12 from Melbourne, Australia and 12 from Athens, Greece) who were required to have visual acuities of 6/7.5 or better in each eye and no reported systemic or eye conditions known to alter vision. All participants were naïve to the exact purpose of the study. All procedures were in accordance with the 1964 Declaration of Helsinki and were approved by our local institutional ethics committees. Participants gave their informed consent prior to inclusion in

Results

Fig. 2 gives reciprobit plots of latency distributions for the three tasks, along with the best fitting simulation of the data, for the two observers with the best (Panel A) and worst (Panel B) fitted distributions overall, as judged by the sum of the KS statistic across tasks. For all datasets for all observers (n = 23 participants), the fitted distributions did not differ significantly from the empirical distributions (p > 0.25 for all KS statistics). Fig. 3 gives the median latencies of all

Discussion

We found no significant difference between saccadic latencies in the step task and control trial latencies in the Wheeless task, nor in the model parameters fitting these two latency distributions. However, latencies in the countermanding task were increased with respect to the step task. LATER modelling indicated that this latency increase was the result of an increase in the threshold the decision unit needed to reach (ST), as well as a decrease in the rate of rise of the decision unit (μ) (

Acknowledgements

Supported by Australian Research Council grant DP120100651 to AJA and RHSC, and FT120100407 to AJA.

Declarations of Interests

None.

Author Contributions

AJA: experimental design, data collection, analysis, & manuscript writing; RHSC: experimental design, & preliminary analysis and manuscript writing; IN: Analysis & manuscript writing; NS: experimental design, data collection, analysis, & manuscript writing.

Data Sharing

Individual participant age, sex, site (Melbourne vs Athens), median latencies, and best fit model parameters are available on Figshare: doi 10.26188/5e3b94cec0a32.

References (46)

  • M. Mahjoob et al.

    Contrast discrimination under task-induced mental load

    Vision Res

    (2019)
  • B.C. Regan et al.

    Luminance noise and the rapid determination of discrimination ellipses in colour deficiency

    Vision Res

    (1994)
  • N. Smyrnis et al.

    Larger variability of saccadic reaction times in schizophrenia patients

    Psychiatry Res

    (2009)
  • N. Amatya et al.

    Differing proportions of ‘express saccade makers’ in different human populations

    Exp Brain Res

    (2011)
  • C.A. Antoniades et al.

    The relationship between abnormalitites of saccadic and manual response times in Parkinson’s disease

    J Parkinsons Dis

    (2013)
  • C.A. Antoniades et al.

    Huntington’s disease: changes in saccades and hand-tapping over 3 years

    J Neurol

    (2010)
  • R.H.S. Carpenter

    Oculomotor procrastination

  • R.H.S. Carpenter

    Movements of the eyes

    (1988)
  • R.H.S. Carpenter

    SPIC: a PC-based system for rapid measurements of saccadic responses

    J Physiol (Proc)

    (1994)
  • R.H.S. Carpenter

    A neural mechanism that randomises behaviour

    J Consciousness Stud

    (1999)
  • Carpenter RHS, Reddi BAJ, Anderson AJ (2009) A simple two-stage model predicts response time distributions. J Physiol...
  • R.H.S. Carpenter et al.

    Neural computation of log likelihood in control of eye movements

    Nature

    (1995)
  • Emeric EE, Brown JW, Boucher L, Carpenter RHS, Hanes DP, Harris R, Logan GD, Mashru RN, Paré M, Pouget P, Stuphorn V,...
  • View full text