PreprintPDF Available

One-shot learning of proactive effect monitoring: Temporally-distinct attention allocation towards future action outcomes

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Cognitive control is essential for adaptive goal-directed action and has been linked to learning. Here, we investigated whether proactive effect monitoring (i.e., a proactive cognitive control process) could be instantiated by one-shot learning. When an action consistently yields the same effect, we form a bi-directional action-effect association. This allows us to both select the appropriate action to achieve a desired effect by anticipating it and to initiate a proactive effect monitoring process that prepares the later comparison of expected and actual effect. Proactive effect monitoring is reflected in anticipatory saccades towards the location of an action's future effect. Importantly, anticipatory saccade latencies also reflect expected effect delays, allowing for an assessment of the quality of proactive effect monitoring. In light of recent studies demonstrating that one-shot learning of control is possible, we assessed whether temporally-distinct (i.e. qualitatively-efficient) proactive effect monitoring could be instantiated by one-shot learning. Participants responded to targets with left/right key presses. Correct left/right responses predictably led to a visual effect after a short/long (200ms/800ms) effect delay. The mapping between responses and their respective effect delays changed every four, eight, or twelve trials (randomly allocated), so that participants could not predict when action-effect delay mappings would switch. Anticipatory saccade latencies were longer for long as compared to short effect delays after a single (re-)learning instance to the extent of participants' individual perceptions of time. This demonstrates one-shot learning of qualitatively-efficient proactive effect monitoring. Thus, anticipatory saccades provide a novel method of directly assessing proactive cognitive control. One-shot learning of proactive effect monitoring 3 Public Significance Statement Our actions often cause predictable consequences in our environment. Through everyday experience, we learn which action causes which consequence. This allows us to anticipate future consequences of our actions and already move our eyes to where these consequences will appear in the future which supports our adaptation to unexpected events. Here, we show that we only need to experience the timing of our actions' consequences once to anticipatorily look towards their future position correspondingly early/late, in a temporally-adapted way. The degree to which we anticipatorily adapt to the timing of our actions' future consequences further depends on individual timing abilities. One-shot learning of proactive effect monitoring 4
Content may be subject to copyright.
One-shot learning of proactive effect monitoring 1
One-shot learning of proactive effect monitoring:
Temporally-distinct attention allocation towards future action outcomes
Florian Gouret & Christina U. Pfeuffer
Cognition, Action, and Sustainability Unit, Department of Psychology, Albert-Ludwigs-
University of Freiburg, Freiburg, Germany
Draft version 20/12/2021. This paper has not yet been peer reviewed. Please do not copy or
cite without author's permission.
Author Note
This research was supported by a grant of the Baden-Württemberg Stiftung awarded to
Christina U. Pfeuffer who is indebted to the Baden-Württemberg Stiftung for the financial
support of this research project by the Eliteprogramm for Postdocs.
We thank Benjamin Gaßmann, Salomé Li Keintzel, Ana-Maria Rosca, Insa Schaffernak,
Jonathan Caspari, and Jonas Plate for their help with data collection.
The data as well as experiment files and syntaxes are publicly available on OSF:
https://osf.io/TB8KY/ - DOI 10.17605/OSF.IO/TB8KY
Word count: 6635
Correspondence:
Florian Gouret
Albert-Ludwigs-Universitaet Freiburg
Cognition, Action, and Sustainability Unit, Department of Psychology
Engelbergerstrasse 41
79085 Freiburg, Germany
Tel. +49-761-203-5689
Email: florian.gouret@psychologie.uni-freiburg.de
One-shot learning of proactive effect monitoring 2
Abstract
Cognitive control is essential for adaptive goal-directed action and has been linked to
learning. Here, we investigated whether proactive effect monitoring (i.e., a proactive
cognitive control process) could be instantiated by one-shot learning. When an action
consistently yields the same effect, we form a bi-directional action-effect association. This
allows us to both select the appropriate action to achieve a desired effect by anticipating it
and to initiate a proactive effect monitoring process that prepares the later comparison of
expected and actual effect. Proactive effect monitoring is reflected in anticipatory saccades
towards the location of an action’s future effect. Importantly, anticipatory saccade latencies
also reflect expected effect delays, allowing for an assessment of the quality of proactive
effect monitoring. In light of recent studies demonstrating that one-shot learning of control is
possible, we assessed whether temporally-distinct (i.e. qualitatively-efficient) proactive effect
monitoring could be instantiated by one-shot learning. Participants responded to targets with
left/right key presses. Correct left/right responses predictably led to a visual effect after a
short/long (200ms/800ms) effect delay. The mapping between responses and their
respective effect delays changed every four, eight, or twelve trials (randomly allocated), so
that participants could not predict when action-effect delay mappings would switch.
Anticipatory saccade latencies were longer for long as compared to short effect delays after
a single (re-)learning instance to the extent of participants’ individual perceptions of time.
This demonstrates one-shot learning of qualitatively-efficient proactive effect monitoring.
Thus, anticipatory saccades provide a novel method of directly assessing proactive cognitive
control.
Keywords: cognitive control; monitoring; associative learning; ideomotor theory; anticipatory
saccades
One-shot learning of proactive effect monitoring 3
Public Significance Statement
Our actions often cause predictable consequences in our environment. Through everyday
experience, we learn which action causes which consequence. This allows us to anticipate
future consequences of our actions and already move our eyes to where these
consequences will appear in the future which supports our adaptation to unexpected events.
Here, we show that we only need to experience the timing of our actions‘ consequences
once to anticipatorily look towards their future position correspondingly early/late, in a
temporally-adapted way. The degree to which we anticipatorily adapt to the timing of our
actions‘ future consequences further depends on individual timing abilities.
One-shot learning of proactive effect monitoring 4
Introduction
Cognitive control refers to the ability of humans to adapt their behaviour based on
changing environmental conditions to achieve their goals (e.g., Atkinson & Shiffrin, 1968;
Bugg, 2008, 2012; Collins & Koechlin, 2012; Egner, 2005, 2007; Norman & Shallice, 1986).
It consists of two mechanisms: Proactive control in preparation of future situations (i.e., goal-
directed and anticipatory) and reactive control in response to already present challenges to
our opportunities for goal-attainment (i.e., stimulus-based; e.g., Braver, 2012). The ability to
anticipate and prepare for changes by monitoring one’s environment (proactive cognitive
control) is especially vital to thrive and successfully act according to our goals (see e.g.,
Braver, 2012; Miller & Cohen, 2001; see Jiang et al., 2014, for a Bayesian perspective).
Thus, monitoring is an important aspect of (proactive) cognitive control when it comes to
enabling adaptive, goal-serving action (see e.g., Krigolson & Holroyd, 2007; Miyake et al.,
2000). Here, we assessed whether proactive monitoring (i.e., a proactive cognitive control
process) could be established by one-shot learning.
Although monitoring processes are essential for understanding the control of human
actions, monitoring has mostly been indirectly assessed via performance differences (see
e.g., Aufschnaiter et al., 2018; Band et al., 2009; Egner, 2007; Yeung et al., 2004, for
monitoring processes in different domains). Interestingly, a new way of assessing proactive
monitoring of future action outcomes (as opposed to reactive monitoring, i.e., attention
allocation towards already present stimuli/effects) and its time course more directly via
anticipatory eye movements was developed by Pfeuffer et al. (2016; see also Gouret &
Pfeuffer, 2021, Pfeuffer et al., under revision/preprint; see e.g., Herwig & Horstmann, 2011;
Huestegge & Kreutzfeldt, 2012; for further evidence that monitoring of future action
outcomes is reflected in eye movements). When actions predictably yield certain sensory
effects, we anticipatorily shift our attention towards the future locations at which we expect
these effects of our actions to occur. These anticipatory saccades occurred without
instruction or (explicit) information regarding action-effect contingencies. Dissociating effects
One-shot learning of proactive effect monitoring 5
on action selection in manual, effect-generating responses and such anticipatory saccades,
Pfeuffer et al. (2016) concluded that spontaneously-occurring, anticipatory
1
(i.e., starting
before effect onset) saccades reflected a proactive effect monitoring process. That is,
participants anticipatorily shifted their attention to future effect locations to prepare the later
comparison of expected and actual effect and adapt accordingly. Similar findings have been
obtained for eye movement in everyday situations. For instance, when making a sandwich or
preparing some tea, that is, when performing action sequences, we already look towards the
next object we plan to interact with while finishing the current task (Land & Hayhoe, 2001;
see e.g. also research on the eye-hand span for examples how eye movements can be
associated with actions: Hayhoe et al., 2003, Land et al., 1999).
In the context of proactive effect monitoring, it is evident how strongly intertwined
cognitive control (here monitoring) and learning are (for evidence that cognitive control and
learning are connected see e.g., Abrahamse et al., 2016; Blais et al., 2007; Braem & Egner,
2018; Bugg & Crump, 2012; Chiu & Egner, 2019; Egner, 2014; Pfeuffer et al., 2019; Verguts
& Notebaert, 2008). When a goal-directed action predictably yields the same effect, a bi-
directional association is formed between this action and the effect (e.g., Dutzi & Hommel,
2008, Elsner & Hommel, 2001; Hommel et al., 2001; Kunde, 2001). Due to these bi-
directional action-effect associations, by anticipating a desired effect, we can select an
appropriate action to produce it (e.g., Hommel, 2009; Hommel et al., 2001; see e.g., Kunde,
2001, for evidence that anticipated future effect influence action selection). This anticipation
1
Please note that we use the term anticipatory to refer to processes based on the
anticipation of one’s actions’ future effects. The anticipation of future effects leads both to a
process of action selection for the effect-generating action (e.g., manual key press; e.g.,
Hommel et al., 2001; Kunde, 2001) and to a process of proactive effect monitoring as
evidenced by anticipatory saccades (e.g., Pfeuffer et al., 2016). Both processes are crucially
influenced by the anticipation of future effects. As such, we consider effect monitoring
processes (i.e., corresponding saccades) occurring prior to effect onset (i.e., during an
anticipatory interval when the effect has not been determined with certainty yet) as based on
effect anticipation even if they occur after effect-generating responses.
One-shot learning of proactive effect monitoring 6
of a future effect is also the basis of the proactive effect monitoring process reflected in
anticipatory saccades (e.g., Pfeuffer et al., 2016) we described.
Interestingly, reactive cognitive control can be instantiated by one-shot learning, that
is, a single exposure to the pairing of stimulus and control state (Whitehead et al., 2020, see
also Brosowsky & Crump, 2018). For instance, task switch (trial n-1 to trial n) costs for a
specific stimulus are reduced when re-encountering a stimulus that has previously once
been presented in a (trial n-1 to trial n) task switch as compared to task repetition trial. Here,
we questioned to what degree one-shot learning of control could also take place for
proactive (proactive effect monitoring) rather than reactive cognitive control.
First evidence that one-shot learning of proactive effect monitoring was possible was
provided by Gouret and Pfeuffer (2021). To assess how many action-effect location (re-
)learning instances were needed for anticipatory saccades towards future effects (i.e.,
proactive effect monitoring) to emerge, they unpredictably manipulated the number of trials
that action-effect position mappings (e.g., left key presseffect on the right) remained
constant. Immediately after experiencing a change in action-effect location mappings (trial 1
of the new sequence), participants started performing significantly more saccades towards
the future effect (effect-congruent) than away from the future effect (effect-incongruent). This
demonstrates that proactive effect monitoring of future effect locations emerges after one-
shot learning of control.
Yet, this finding can only tell us that a proactive effect monitoring process is active.
To efficiently monitor, however, our attention needs to be shifted towards future effect
locations at the right time (neither too early nor too late). Thus, we assumed that cognitive
control is not an on or off process, but might differ in efficiency (i.e., quality) regarding, for
instance, temporal aspects of future effects. We propose that the quality of proactive effect
monitoring shows in earlier/later saccades depending on the expected delay of future effects
(see Pfeuffer et al., in preparation/2019, for first evidence that effect delay affects proactive
effect monitoring). That is, an efficient proactive effect monitoring process is triggered
One-shot learning of proactive effect monitoring 7
temporally-distinctively at the appropriate time. Here, we investigated whether this
temporally-distinct, that is, qualitatively-efficient proactive effect monitoring could be
established by one-shot learning.
Interestingly, a study of Pfeuffer et al. (in preparation/2019) manipulating effect delay
with a different focus found that even small differences in predictable effect delays (200 ms
vs. 800 ms) influenced proactive effect monitoring (see Dignath et al., 2014, Dignath &
Janczyk, 2017; Elsner & Hommel, 2004; Haering & Kiesel, 2014; Janczyk et al., 2017;
Moore et al., 2008; Riechelmann et al., 2017; Shin & Proctor, 2012; Wirth et al., 2015, for
evidence that anticipated effect delay also impacts on action selection). Participants’ first
saccade towards the future effect per trial was performed significantly later when a long
effect delay rather than a short effect delay preceded effects. That is, Pfeuffer et al. (in
preparation/2019) provided first evidence for the idea that anticipatory saccades could also
reflect qualitative aspects of proactive effect monitoring. However, there, it could not be
assessed how proactive effect monitoring was related to learning and after how many
learning instances such temporally-distinct proactive effect monitoring emerged.
Gouret and Pfeuffer (2021) showed that proactive effect monitoring (at least in terms
of an on/off process) could be established by one-shot learning. Here, we built on these
findings and investigated whether temporally-distinct (i.e., qualitatively-efficient) proactive
effect monitoring that takes the timing of future effects into account could be established in a
single (re-)learning instance as well. To address this question, we conducted an experiment
based on the paradigm developed by Gouret and Pfeuffer (2021). Participants performed
forced-choice left/right key presses. Each correct left/right response led to a visual effect
after a short (200 ms) or long (800 ms) effect delay. The action-effect delay mappings
remained stable for randomly intermixed sequences of four, eight, or twelve trials and then
reversed, making switches in action-effect delay mappings unpredictable.
Our aim was to show that qualitatively-efficient proactive effect monitoring taking
effect timing into account can be established by one-shot learning of control. If participants
One-shot learning of proactive effect monitoring 8
were at all able to proactively monitor effects within a few trials, we expected the latency of
participants’ first anticipatory saccade towards the future effect to be shorter for the short
(200 ms) as compared to the long (800 ms) effect delay. If one-shot learning of temporally-
distinct proactive effect monitoring (i.e., qualitively-efficient proactive control) were possible,
this should be the case after a single action-effect delay (re-)learning instance.
Importantly, the performance of anticipatory saccades at the appropriate time might
rely on participants’ perception of time or ability to estimate time (see e.g. Bryce & Bratzke,
2017; Eagleman, 2008; Hayashi et al., 2014, for further information on individual time
perception). Thus, to account for a possible influence of individual time perception, in a
separate phase at the end of the experiment, we assessed participants’ individual time
perception. We expected that participants with good rather than poor ability to distinguish the
different effect delays might also show larger latency differences between short and long
effect delays in anticipatory saccades.
Method
Participants
A prior study assessing the impact of effect delays on anticipatory saccade latencies
when action-effect delay mappings were stable for more than 100 trials (Pfeuffer et al., in
preparation/2019) reported an effect size of dz = 2.26 for the difference in saccade latencies
between the short and long effect delay (200 vs. 800 ms). Assuming that effect size as a
maximum, we calculated with an effect size four times smaller, dz = 0.57, for the current
experiment in which action-effect delay mappings frequently switched. An a priori sample
size estimation = .05, 1-β = .80) suggested that at least 21 participants were required to
find an effect of this size (GPower, Faul et al., 2007).
Twenty-five participants were recruited for the experiment (9 male, 16 female, mean
age = 21.6 years, SD = 5.4 years, 1 left-handed, 8 left eye dominant, 16 Psychology
students, 9 students of other subjects). All participants had normal or corrected-to-normal
One-shot learning of proactive effect monitoring 9
vision and were naive to the purpose of the experiment. They provided written informed
consent prior to their participation and received either course credit or 14€.
The study was conducted in agreement with the Declaration of Helsinki (World
Medical Association, 2013) and the guidelines set by the local ethics committee.
Stimuli and Apparatus
The experiment was conducted in a dark and sound attenuated laboratory room.
Participants sat approximately 60 cm from a 24” LCD screen (1920 pixels x 1080 pixels, 144
Hz) with their index fingers on a left and right key (distance 13.5 cm), respectively. The
experiment was run via EPrime (PST, Sharpsburg, USA; version 2.0.10.356). During the
entire experiment, the background of the screen was black.
Eye movements were tracked with an EyeLink 1000 Plus Desktop Mount (SR
Research Ltd., Ontario, Canada). Corneal reflection and pupil diameter were measured via
an infrared camera and eye movements (dominant eye) were sampled at 1000 Hz with a
spatial resolution of 0.01° visual angle. At the beginning of each block, calibration and
validation were performed.
Design and procedure
The experiment consisted of a practice (24 trials) and 13 blocks (72 trials each) of the
main experiment. During the practice in which no effects were presented yet, participants
were allowed to ask questions to ensure they understood how to correctly perform the task.
The practice was then followed by the main experiment which consisted of two phases: The
experimental phase (12 blocks) and a subsequent time estimation phase (1 block).
Participants’ eyes were only tracked and their manual and saccadic performance assessed
during the experimental phase. During the time estimation phase, participants’ ability to
judge the duration of the effect delays was assessed. At the end of each block, participants
received feedback about their performance (i.e., number of errors, omitted responses, and
One-shot learning of proactive effect monitoring 10
premature responses) and could take a self-paced break. They were also reminded to
respond as fast and accurately as possible.
Each experimental trial began with the presentation of a white fixation cross in the
middle of the screen (intertrial interval, ITI; see Figure 1 for the trial structure). The duration
of the intertrial interval varied between 1200 ms and 1500 ms to prevent temporal
anticipations of the target stimuli. The fixation cross was then replaced by a target displayed
for 100 ms (height: 0.7°). In the first trial of a block, the target was a white arrow pointing
either to the left or to the right; the arrow’s direction indicated the first key to press. In the
following trials, the target was either an “=” or an “x”. The “=” target indicated that
participants were to press the same key as in the previous trial (same response), whereas
the “x” target indicated that participants were to press the opposite key as compared to the
previous trial (opposite response). The reference was always the correct response on the
previous trial. Once the target disappeared, a blank screen was displayed (for a maximum
duration of 1400 ms or until response) and participants could respond for a maximum of
1500 ms from target onset.
Correct responses were followed by a blank screen action-effect interval lasting
either 200 ms or 800 ms (effect delay). There was one specific effect delay per response
(e.g., left response200 ms, right response800 ms). The action-effect delay mappings
were constant for sequences of either four, eight, or twelve trials. At the end of each
sequence, the action-effect delay mappings switched. Per sequence, half of the trials
required right and left manual responses. Each sequence length occurred equally often
throughout the experiment and per block. There were three sequences of four, eight, and
twelve trials each per block and sequences were presented in random order. Thus, switches
in action-effect delay mappings were unpredictable.
After the action-effect interval, an effect appeared and remained on screen for 500
ms. Effects were orange or blue circles (diameter: 1.3°) appearing at 12.6° to the left/right of
the screen centre and moving further in the respective direction (additional movement: 6.6°,
One-shot learning of proactive effect monitoring 11
end position: 19.1°). Effects always appeared at positions spatially-compatible with the
manual response (e.g., left manual response -> effect on the left). The colour of the effect
was contingent on the response (e.g., orange for the left and blue for the right response).
Then, the next trial followed. The starting action-effect delay mappings at the beginning of
the experiment and the action-effect colour mappings were counterbalanced across
participants.
In case of an incorrect, premature, or omitted response, no effect was presented,
instead feedback was displayed in red in the centre of the screen (“zu früh!”/”too early!” for
premature responses, “Fehler!”/”error!” for incorrect responses, and “zu langsam!”/”too
slow!” for response omissions; duration: 1000 ms), and the trial was aborted.
Participants were instructed to respond as fast and accurately as possible. In the
experimental phase, eye movements occurring between target offset and effect onset were
assessed. No information except that their responses would be followed by coloured circles
was provided to participants in advance. Moreover, participants were not informed about
effect delays, changes in action-effect delay mapping, or sequences. Furthermore, no
information or instruction regarding eye movements was given to participants to avoid
biases. Therefore, any saccade performed during the anticipatory interval between target
offset and effect onset can be considered as spontaneous and uninstructed.
After twelve experimental blocks, participants answered post-experiment questions to
determine whether they had noticed any temporal regularities and were then introduced to
the last phase of the experiment, that is, the time estimation phase. There, the trial structure
was exactly the same as during the experimental phase until the effect was presented. After
the effect was displayed, a screen appeared which asked participants to press and hold the
‘SPACE’ key for as long as they thought the action-effect interval (i.e., effect delay) between
their last manual response and the appearance of the effect had lasted (see e.g., Bratzke et
al., 2017; Bryce & Bratzke, 2017, for a comparison of several time estimation methods). No
eye movements were recorded during the time estimation phase.
One-shot learning of proactive effect monitoring 12
Results
Pre-processing and Trial Inclusions
Trials with premature (<0.1%), omitted (0.7%), or erroneous (4.7%) responses were
excluded from all analyses. Moreover, participants were not informed about changes in
action-effect delay mapping. Thus, performance in the first trial of a new sequence was not
informative, as participants had not yet experienced the change. We therefore excluded
sequence trial one from all analyses. Furthermore, as action-effect delay mappings
predictably switched between blocks, the first sequence of each block was also excluded
from all analyses.
Saccades were detected using the standard settings of SR Research’s software, that
is,according to a combined velocity (30°/s), motion (0.1°), and acceleration (8000°/s2)
threshold. Only saccades occurring in experimental blocks during the anticipatory interval
between target offset
2
and effect onset were considered (correctly-responded to sequence
trials 2-12 only, sequence trial 1 was excluded from all analyses). Furthermore, only
saccades that crossed at least horizontally (18,070 of 27,065 saccades) and saccades
occurring in trials in which the first saccade after target offset started within ± 1.0°
horizontally around screen centre (i.e., around the target location, suggesting that the target
was processed; 16,070 of 27,065 saccades) were considered. In total, 10,082 saccades
fulfilled all criteria and were included in our analyses (37.3% of all saccades; see Gouret &
Pfeuffer, 2021, for the idea that the volatility of action-effect mappings affects the proportion
of large-amplitude saccades). All saccades that fulfilled these criteria were considered when
assessing the relative frequency of saccades towards the future effect (saccade-effect
congruency analysis; i.e., also multiple saccades per trial). For the analysis of saccade
2
As the visual target could have affected eye movements, anticipatory saccades were only
assessed between target offset and effect onset (anticipatory interval). Note, however, that
the time reference for saccade latencies, like manual RTs, was target onset to facilitate
comparisons.
One-shot learning of proactive effect monitoring 13
latencies, only participants’ first saccade towards the future effect (effect-congruent) per trial
was considered (in rare cases where multiple saccades per trial occurred, only 4.0% of trials
with saccades).
Analyses were performed using Visual Code version 1.55.2 for R version 4.0.5 (R
Core Team, 2020). Linear mixed models (LMMs) were used to assess saccade latency (and
manual reaction time; see Appendix B) and generalized linear mixed model (GLMMs) were
used to assess saccade-effect congruency (and errors; see Appendix B). (G)LMM analyses
inluded effect delay (200 ms = -1 vs. 800 ms = 1; effect coding) and response occurrence.
Response occurrence corresponded to the Xth time participants correctly performed the
respective response in the respective sequence leading to an effect (response occurrence 1
= 0 prior experiences of this effect delay following this response in the current sequence;
response occurrences: 1 – 6 at maximum for 12 trial sequences).
3
Theoretically (not
considering errors or response omissions), per person, 117 trials could be considered for
response occurrence 1, 234 for response occurrence 2, 156 for response occurrence 3 and
4 have each and 78 for response occurrence 5 and 6 each. To fit (G)LMMs, response
occurrences were recoded as follows: 1 = -2.5, 2 = -1.5, 3 = -0.5, 4 = 0.5, 5 = 1.5, 6 = 2.5.
Simple coding was used to separately compare the reference, response occurrence 1, to
response occurrences 2-6 while taking the grand mean into account. (G)LMMs included
3
Please note that we coded the trials in the sequence in this way, as (successful) response
occurrences, as participants could only learn about response-specific effect delays on trials
on which they correctly responded and produced a corresponding effect. That is, the trial
position in the sequence itself is relatively uninformative about the number of previous
learning instances a person has previously had in the corresponding sequence.
Nevertheless, to display saccade patterns in the time course of trial sequences, we
additionally present the distribution of saccades per trial in the sequence (2-12) in the
Supplementary Material.
Moreover, please note that saccades were not instructed and therefore saccades that
fulfilled our minimum amplitude criterion only occurred on a varying portion of trials per
participant. Furthermore, saccades from trials with erroneous responses or response
omissions could not be considered. Therefore, an assessment of, for instance, saccade
latency differences between the last corresponding trial of the previous sequence and the
trials of the current sequence was, unfortunately, not possible due to the substantial number
of trials without saccades.
One-shot learning of proactive effect monitoring 14
response occurrence and effect delay as well as their interaction as fixed effects. In addition,
participants’ individual, z-standardized time estimation score (continuous predictor; see
following section for computation details) and its interactions with the other predictors were
included as fixed effects. More detailed model specifications and descriptions of the
packages used can be found in Appendix A. Tables with detailed model outputs including
information on the random effects structures included in the final models can be found in the
Supplementary Materials.
Time estimation and post-experiment reports
In the time estimation phase (block 13), at the end of each correct trial that produced
an effect, participants were asked to indicate the duration of the action-effect interval (i.e.,
the effect delay) by pressing and holding the SPACE key. We used their time estimations to
compute a score reflecting their individual time estimation ability. Participants’ time
estimation score reflected the ratio between participants’ time estimations and the actual
durations. First, a trial score was determined by dividing the respective estimated duration
by the actual duration (200 ms / 800 ms) and then subtracting 1 (re-centering around 0; i.e.,
a score of 0 indicates perfect time estimation, scores below/above 0 indicate under-
/overestimation) for each individual trial. Then, each participants’ time estimation score was
determined by computing the average of their respective trial scores (
, with n being the number of trials; M =
-0.51, SD = 0.61). Time estimation scores were then z-standardized across participants
before they were submitted to the (G)LMMs. Overall, the majority of participants tended to
underestimate effect delays (M before z-standardization = -0.51) and only very few
participants estimated almost correctly or showed overestimations (effect delay 200 ms: M =
187 ms, SD = 85.5 ms; effect delay 800 ms: M = 445 ms, SD = 220.5 ms). That is, on
average, participants’ time estimations were not distributed around the actual effect delays,
but around substantially shorter delays. Thus, low z-standardized time estimation scores
One-shot learning of proactive effect monitoring 15
indicated temporal underestimations, whereas high scores indicated accurate time
estimation ability to temporal overestimations.
Post-experiment questions
4
showed that 10 participants (of 25) correctly indicated
that there were two effect delays. None of the participants reported noticing any regularities
in the effect delays. All participants remained naïve to the purpose of the experiment.
Manual responses
Overall, participants mean error rate was 4.7% and the mean reaction time (RT) was
598 ms. Parallel manual response analyses are reported for the sake of completeness, but
not considered in the main text as our hypotheses only concerned anticipatory saccades.
Anticipatory saccades
The total number of saccades included for the relative saccade frequency analysis
(both congruent and incongruent saccades fulfilling all the inclusion criteria: crossing at least
horizontally, first horizontal eye position at target offset at target center ± ) was 10,082.
That is, 1,472 saccades for response occurrence 1 (14.6% of all saccades), 2,923 saccades
for response occurrence 2 (29.0% of all saccades), 1,979 saccades for response occurrence
3 (19.6% of all saccades), 1,877 saccades for response occurrence 4 (18.6% of all
saccades), 1,028 saccades for response occurrence 5 (10.2% of all saccades); 803
saccades for response occurrence 6 (8.0% of all saccades), respectively, were considered.
For the analysis of saccade latencies, only the first effect-congruent (i.e., in the
direction of the future effect) saccade per trial was considered. Thus, the number of
saccades considered for the saccade latency analysis was lower. In total, participants
performed 8609 first effect-congruent saccades. Respectively, participants in total performed
1,271 saccades for response occurrence 1 (14.8% of all saccades considered for the
saccade latency analysis), 2,507 for response occurrence 2 (29.1%), 1,713 for response
4
Post-experiment questionnaires can be found in OSF (https://osf.io/TB8KY/)
One-shot learning of proactive effect monitoring 16
occurrence 3 (19.9%), 1,580 for response occurrence 4 (18.4%), 855 for response
occurrence 5 (9.9% of all saccades), and 683 for response occurrence 6 (7.9%).
Relative saccade frequencies. The saccade effect-congruency (SEC) score
indicates the relative frequency of saccades that were directed towards the future effect
location (towards effect location = effect-congruent; towards opposite, non-effect location =
effect-incongruent) as compared to all saccades performed by a participant in the respective
condition (SEC = ). A value of 50% represents the
chance level. Values above 50% indicate that participants correctly anticipated their action’s
future effect’s location and moved their eyes towards it in anticipation. That is, they
proactively monitored their actions’ future effects. When multiple saccades that fulfilled the
inclusion criteria occurred in a trial, each saccade was individually marked as effect-
congruent/effect-incongruent based on its direction and all saccades that fulfilled the basic
inclusion criteria were included in the analysis of relative saccade frequencies irrespective of
whether they were the first saccade of a trial or a later saccade (only 4.0% of trials contained
more than one saccade). The direction of a saccade was determined by computing the
difference between its starting position and end position (see Figure 1 in the Supplementary
Materials for the distribution of saccade end positions/saccadic gain across conditions).
A one-sample t-test showed that participants mean SEC scores (M = 80.8%, SD =
30%) were significantly greater than 50%, t (24) = 7.19, p < .001, d = 1.44 (short effect-
delay, M = 81.2%, SD = 30%, t (24) = 7.49, p < .001, d = 1.50; long effect delay, M = 80.2%,
SD = 29%, , t (24) = 7.00, p < .001, d = 1.40).
Action-effect location mappings remained constant throughout the experiment. Thus,
differences in SEC scores between conditions would indicate an influence of anticipated
effect delays on the direction of anticipatory saccades (towards/away from future effect). No
difference between conditions would suggest that effect location and effect delay affect
separate parameters of anticipatory saccades, direction and latency, respectively (see
One-shot learning of proactive effect monitoring 17
Gouret & Pfeuffer, 2021, for the opposite dissociation: Manipulations of effect location
affecting SEC scores, but not anticipatory saccade latencies).
The GLMM fitting saccade-effect congruency (0 = saccade effect-incongruent/away
from future effect, 1 = saccade effect-congruent/towards future effect) included only
participant intercepts as random effects (see Figure 2 for the GLMM results). The main
effect of time estimation, z = 1.90, p = .057, OR = 1.85, and all other fixed effects as well as
their interactions failed to reach significance, zs 1.90, ps .097 (see Table 1 in
Supplementary Materials). That is, saccade-effect congruency (SEC) scores were not
significantly affected by response occurrence, effect delay, individual time estimation ability,
or their interactions.
Saccade latency. For the latency analysis
5
, we only considered participants’ first
effect-congruent saccade per trial. The LMM on the latency of participants’ first effect-
congruent saccade per trial included participant intercepts and by-participant random slopes
for effect delay as random effects. Most importantly, the main effect of effect delay reached
significance, t = -4.92, p < .001,
6
= 98 ms (see Figure 3). That is, participants performed
their first effect-congruent saccade later for long rather than short effect delays. Moreover,
effect delay and time estimation score significantly interacted, t = 2.29, p = .032, = 46 ms.
A post-hoc test showed that the slopes of participants’ time estimation scores significantly
differed between effect delays, t = 2.29, p = .022, = 92 ms, with steeper slopes for the
5
We additionally assessed saccade-manual latency differences between saccade latencies
and manual RTs on corresponding trials (see Appendix C) to account for possible influences
of manual responses on anticipatory saccades. As results were equivalent, we only report
these analyses in the Appendix.
6
The notation is commonly used to indicate the size of an effect in LMMs when the
prevalent effect coding is used. In this case, to compare two levels of a predictor, one
condition is compared to the grand mean. The estimate of the size of this effect (β) is
reported in the model table. The actual size of the effect, that is, the difference between the
two levels of the predictor, in this case corresponds to 2β. This is how we coded contrasts
for effect delay. For the simple coding of response occurrence, the estimate (β) reflects the
full effect. For the sake of consistency, we will speak of whenever we report LMM effect
sizes. Note however, that the reported values correspond to β in case of the predictor
response occurrence in accordance with the contrast coding we chose.
One-shot learning of proactive effect monitoring 18
long as compared to the short effect delay. That is, participants showed larger latency
differences between effect delay conditions the more they tended to underestimate the
duration of effect delays. Additionally, latencies significantly differed between response
occurrence 1 and 2, t = 2.08, p = .038, = 27 ms, and response occurrence 1 and 3, t =
3.41, p = .001, = 46 ms. Participants performed their first effect-congruent saccade earlier
on response occurrence 1 as compared to response occurrence 2 and 3 (see Figure 2 in
Supplementary Material for an illustration of saccade latency distributions per trial position in
a sequence). The interaction between effect delay and the comparison between response
occurrence 1 vs. 5 showed a non-significant trend, t = -1.89, p = .059, = 31 ms. Other
effects did not reach significance, ts1.70, ps.090 (see Table 2 in Supplementary
Materials)
7
.
Discussion
In the present study, we investigated whether action-effect delay associations that
led to temporally-distinct proactive effect monitoring (i.e., a qualitatively-efficient proactive
control process) could be established by one-shot learning. Participants’ correct responses
were followed by a visual effect after a short (200 ms) or long (800 ms) effect delay. Action-
effect delay mappings (e.g., left response -> 200 ms effect delay, right response -> 800 ms
effect delay) were predictable in short trial sequences. Unbeknownst to participants,
mappings between responses and their effects’ delays unpredictably reversed after
sequences of four, eight, or twelve trials (randomly intermixed). Thus, we were able to
examine how fast action-effect delay associations were adapted leading to earlier/later (i.e.,
temporally-distinct = qualitatively-efficient) proactive effect monitoring (i.e., anticipatory
saccades) for responses associated with a short/long effect delay.
7
Note that, the pattern of results for response occurrence and effect delay did not
significantly differ when fitting the same model without considering participants’ time
estimation scores.
One-shot learning of proactive effect monitoring 19
First, participants were unable to report temporal regularities between actions and
effect delays, suggesting that our findings did not rely on participants’ explicit knowledge of
sequences or action-effect delay mappings. Moreover, in line with prior research (see e.g.,
Bryce & Bratzke, 2017; Eagleman, 2008; Hayashi et al., 2014), time estimation scores
showed sufficient variability (z values ranging from -1.47 to 2.24) to (potentially) account for
variance in participants’ eye movements. Importantly, across effect delays, the majority of
participants tended to underestimate effect delays (low z-standardized time estimation
scores).
Participants received no information regarding effects and no instructions regarding
their eye movements. Nevertheless, they looked significantly more often towards their
actions’ effects future locations than in the opposite direction (SEC effect). Thus, we
replicated prior studies showing that participants anticipate their actions’ effects and
proactively monitor their future locations (Gouret & Pfeuffer, 2021; Pfeuffer et al., 2016,
under revision/preprint).
We observed that the first effect-congruent saccade per trial was performed
significantly later when a long rather than a short effect delay preceded an effect. This was
the case from response occurrence 1, without a single prior action-effect delay (re-)learning
instance (but with 1 relearning instance of the opposite response in sequence trial 1;
interestingly, this pattern of results is not visible in manual action selection within 6 response
occurrences, see Appendix B). Thus, participants showed temporally-distinct (i.e.,
qualitatively-efficient) proactive effect monitoring that took the timing of future effects into
account after one-shot learning (see Pfeuffer et al., in preparation/2019, for saccade latency
differences based on effect delay after more than 100 trials). That is, for the first time, we
demonstrate one-shot learning of control for a proactive control process (see Whitehead et
al., 2020, for prior evidence regarding reactive control). More precisely, we show one-shot
learning of proactive control for a temporally-distinct, that is, qualitatively-efficient process of
proactive effect monitoring.
One-shot learning of proactive effect monitoring 20
Interestingly, models of saccadic decisions in space and time like LATEST (Tatler et
al., 2017) suggest that saccades are performed once sufficient evidence to support a
decision has accumulated. In the present study, evidence for the future location of the effect
should have accumulated at about the same time irrespective of the effect delay following
participants’ manual responses. Nevertheless, in the present study, saccades were delayed
when effect delays were longer. Thus, the present findings highlight that saccade latencies
are not only determined by the time of evidence accumulation, but also critically affected by
proactive cognitive control processes like proactive effect monitoring. This indicates that
saccadic decision making additionally takes into account when participants should shift their
attention in order to ideally perceive and process upcoming effects. Thus, our findings match
recent research suggesting that temporal predictability and the timing of attentional shifts
support stimulus-based actions (e.g., Carrasco, 2014; Denison et al., 2019; Pfeuffer et al.,
2020, Thomaschke & Dreisbach, 2015; van Ede et al., 2018).
In addition, our results also illustrate how participants were able to adapt and (re-
)learn their already existing action-effect associations. Right after experiencing a switch in
the action-effect delay mapping for one response, participants already showed latency
differences between effect delays for the first response occurrence of the respective other
response they had not performed yet in the current sequence (i.e., a response whose effect
delay they had not yet experienced in the current sequence; see Gouret & Pfeuffer, 2021, for
the equivalent pattern in SEC effects due to switches in action-effect location mappings).
Thus, participants inferred the future effect delay for the respective other response without a
single re-learning instance of this response. This implies that participants either restructured
both opposing action-effect delay mappings when they noticed a switch in one of the
mappings or that participants retained all four possible mappings, but only held the currently
applicable mappings in an active state. Both of these possibilities would suggest a top-down
process, one restructuring action-effect delay associations and one (dis-)inhibiting action-
One-shot learning of proactive effect monitoring 21
effect delay associations based on prior experience. Interestingly, Gouret & Pfeuffer (2021)
reported similar results for effect location.
In the present study, effect locations were perfectly predictable and remained
constant throughout the experiment. Correspondingly, we did not observe a change in the
SEC effect across sequence trials which was the case for more volatile settings, when
action-effect location mappings frequently switched in the study of Gouret and Pfeuffer
(2021). SEC effects quickly reached an asymptote which was, however, lower in these more
volatile settings than in the present study or prior studies (Pfeuffer et al., 2016, under
revision/preprint). Double dissociating Gouret and Pfeuffer (2021), we found an impact on
saccade latencies due to action-effect delay mappings, but no impact on SEC scores. This
suggests that effect location and time are monitored separately. This finding is also in line
with theories of eye movement generation (e.g., Findlay & Walker, 1999) which also
differentiate between a where (location, here effect location) and when (time, here effect
delay) aspect of eye movements. The assumption of separate monitoring for location and
time is also in line with the idea that actions become independently bi-directionally
associated with different effect features (e.g., locations and delays; see e.g., Dignath &
Janczyk, 2017; Janczyk et al., 2016; Riechelmann et al., 2017), suggesting that the
anticipation and proactive monitoring of the respective effect features might also occur
independently. Nevertheless, fundamental mechanisms underlying proactive monitoring of
different effect features might be similar. This notion is supported by the finding that both
effect location (Gouret & Pfeuffer, 2021) and effect delay can be (re-)learned and monitored
within one trial and can even be inferred for the respective other response.
Interestingly, contrary to our predictions, saccade latency differences between short
and long anticipated effect delays did not increase with the accuracy of individual time
estimations but with the degree of temporal underestimation (i.e., with the inaccuracy of
individual time estimations). We speculate that participants who underestimated effect
delays more strongly perceived a stronger need for temporally-accurate monitoring
One-shot learning of proactive effect monitoring 22
(especially for the short effect delay) to proactively shift their attention towards future effects
in time. This suggests that the quality of proactive effect monitoring regarding effect features
(here effect delay) is sensitive to participants’ individual sensory perceptions of this effect
feature (here time perception). This opens interesting avenues for investigating how the
quality of proactive cognitive control is linked to individual perceptions in corresponding
sensory domains (see e.g., Pallensen et al., 2010, for a link between cognitive control and
auditory working memory; see e.g., Adrover-Roig & Barceló, 2010; Vanneste et al., 2015, for
individual differences in aging and cognitive control). It remains to be examined in detail
whether the same holds true for cognitive control in cases other than monitoring.
Conclusion
Our findings show that one-shot (re-)learning of action-effect delay associations was
sufficient to establish temporally-distinct, that is, qualitatively-efficient proactive effect
monitoring. This establishes anticipatory saccades as a suitable measure of the quality (in
terms of timing efficiency) of proactive effect monitoring. We thus introduce a novel method
of directly assessing the learning of proactive cognitive control and further elucidate the
interdependence of learning and (proactive) cognitive control in goal-directed action based
on the example of proactive effect monitoring. Importantly, the quality (timing efficiency) of
cognitive control also depended on participants’ individual perceptions of time. These effects
were dissociable from action selection processes or the quantity of anticipatory saccades
towards future effect locations.
One-shot learning of proactive effect monitoring 23
Declarations
Funding
This research was supported by a grant of the Baden-Württemberg Stiftung awarded to
Christina Pfeuffer who is indebted to the Baden-Württemberg Stiftung for the financial
support of this research project by the Eliteprogramm for Postdocs
Conflicts of interest/Competing interests
The authors have no conflicts of interest to declare that are relevant to the content of this
article.
Ethics approval
This study was performed in line with the principles of the Declaration of Helsinki and the
University of Freiburg.
Consent to participate
Informed consent was obtained from all individual participants included in the study.
Consent for publication
The authors affirm that human research participants provided informed consent for the
publication of their anonymous data.
Availability of data and material
The data of the reported experiments are available via the Open Science Framework:
https://osf.io/TB8KY/ - DOI 10.17605/OSF.IO/TB8KY
Code availability
Experiment files and syntaxes are available via the Open Science Framework:
https://osf.io/TB8KY/ - DOI 10.17605/OSF.IO/TB8KY
One-shot learning of proactive effect monitoring 24
Authors' contributions
All authors jointly designed the experiments, conducted and analyzed the experiments.
Florian Gouret wrote a first draft. All authors then contributed to adapting and editing the
manuscript.
One-shot learning of proactive effect monitoring 25
References
Abrahamse, E., Braem, S., Notebaert, W., & Verguts, T. (2016). Grounding cognitive control
in associative learning. Psychological Bulletin, 142(7), 693728.
https://doi.org/10.1037/bul0000047
Adrover-Roig, D., & Barceló, F. (2010). Individual differences in aging and cognitive control
modulate the neural indexes of context updating and maintenance during task
switching. Cortex, 46(4), 434450. https://doi.org/10.1016/j.cortex.2009.09.012
Atkinson, R. C., & Shiffrin, R. M. (1968). Human Memory: A Proposed System and its
Control Processes. Psychology of Learning and Motivation, 2, 89195.
https://doi.org/10.1016/s0079-7421(08)60422-3
Aufschnaiter, S., Kiesel, A., Dreisbach, G., Wenke, D., & Thomaschke, R. (2018). Time-
based expectancy in temporally structured task switching. Journal of Experimental
Psychology: Human Perception and Performance, 44(6), 856870.
https://doi.org/10.1037/xhp0000494
Band, G. P. H., van Steenbergen, H., Ridderinkhof, K. R., Falkenstein, M., & Hommel, B.
(2009). Action-effect negativity: Irrelevant action effects are monitored like relevant
feedback. Biological Psychology, 82(3), 211218.
https://doi.org/10.1016/j.biopsycho.2009.06.011
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models
Usinglme4. Journal of Statistical Software, 67(1).
https://doi.org/10.18637/jss.v067.i01
Blais, C., Robidoux, S., Risko, E. F., & Besner, D. (2007). Item-specific adaptation and the
conflict-monitoring hypothesis: A computational model. Psychological
Review, 114(4), 10761086. https://doi.org/10.1037/0033-295x.114.4.1076
One-shot learning of proactive effect monitoring 26
Braem, S., & Egner, T. (2018). Getting a Grip on Cognitive Flexibility. Current Directions in
Psychological Science, 27(6), 470476.
https://doi.org/10.1177/0963721418787475
Braver, T. S. (2012). The variable nature of cognitive control: a dual mechanisms
framework. Trends in Cognitive Sciences, 16(2), 106113.
https://doi.org/10.1016/j.tics.2011.12.010
Bratzke, D., Birngruber, T., Durst, M., & Schröter, H. (2017). Filled and empty motor
reproductions of filled and empty intervals: Is there also a filled - reproduction
illusion? Attention, Perception, & Psychophysics, 79(7), 21432152.
https://doi.org/10.3758/s13414-017-1370-1
Brosowsky, N. P., & Crump, M. J. C. (2018). Memory-guided selective attention: Single
experiences with conflict have long-lasting effects on cognitive control. Journal of
Experimental Psychology: General, 147(8), 11341153.
https://doi.org/10.1037/xge0000431
Bryce, D., & Bratzke, D. (2017). Are participants’ reports of their own reaction times reliable?
Re-examining introspective limitations in active and passive dual-task
paradigms. Acta Psychologica, 172, 1–9.
https://doi.org/10.1016/j.actpsy.2016.10.007
Bugg, J. M., Jacoby, L. L., & Toth, J. P. (2008). Multiple levels of control in the Stroop
task. Memory & Cognition, 36(8), 14841494. https://doi.org/10.3758/mc.36.8.1484
Bugg, J. M. (2012). Dissociating Levels of Cognitive Control. Current Directions in
Psychological Science, 21(5), 302309. https://doi.org/10.1177/0963721412453586
Bugg, J. M., & Crump, M. J. C. (2012). In Support of a Distinction between Voluntary and
Stimulus-Driven Control: A Review of the Literature on Proportion Congruent
effects. Frontiers in Psychology, 3. https://doi.org/10.3389/fpsyg.2012.00367
One-shot learning of proactive effect monitoring 27
Butz, M. V., Sigaud, O., & Gérard, P. (Eds.). (2004). Anticipatory behavior in adaptive
learning systems: Foundations, theories, and systems (Vol. 2684). Springer.
Carrasco, M. (2014). Spatial Covert Attention. In A. C. (Kia) Nobre & S. Kastner (Eds.),
Oxford Handbooks Online. Oxford University Press.
https://doi.org/10.1093/oxfordhb/9780199675111.013.004
Chiu, Y.-C., & Egner, T. (2019). Cortical and subcortical contributions to context-control
learning. Neuroscience & Biobehavioral Reviews, 99, 3341.
https://doi.org/10.1016/j.neubiorev.2019.01.019
Collins, A., & Koechlin, E. (2012). Reasoning, learning, and creativity: frontal lobe function
and human decision-making. PLoS Biology, 10(3), e1001293.
https://doi.org/10.1371/journal.pbio.1001293
Denison, R. N., Yuval-Greenberg, S., & Carrasco, M. (2018). Directing Voluntary Temporal
Attention Increases Fixational Stability. The Journal of Neuroscience, 39(2), 353
363. https://doi.org/10.1523/jneurosci.1926-18.2018
Dignath, D., & Janczyk, M. (2017). Anticipation of delayed action-effects: learning when an
effect occurs, without knowing what this effect will be. Psychological
Research, 81(5), 10721083. https://doi.org/10.1007/s00426-016-0797-7
Dignath, D., Pfister, R., Eder, A. B., Kiesel, A., & Kunde, W. (2014). Representing the
hyphen in actioneffect associations: Automatic acquisition and bidirectional
retrieval of actioneffect intervals. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 40(6), 17011712. https://doi.org/10.1037/xlm0000022
Dutzi, I. B., & Hommel, B. (2008). The microgenesis of action-effect binding. Psychological
Research, 73(3), 425435. https://doi.org/10.1007/s00426-008-0161-7
Eagleman, D. M. (2008). Human time perception and its illusions. Current Opinion in
Neurobiology, 18(2), 131–136. https://doi.org/10.1016/j.conb.2008.06.002
One-shot learning of proactive effect monitoring 28
Egner, T., & Hirsch, J. (2005). Cognitive control mechanisms resolve conflict through cortical
amplification of task-relevant information. Nature Neuroscience, 8(12), 17841790.
https://doi.org/10.1038/nn1594
Egner, T. (2007). Congruency sequence effects and cognitive control. Cognitive, Affective, &
Behavioral Neuroscience, 7(4), 380390. https://doi.org/10.3758/cabn.7.4.380
Egner, T. (2014). Creatures of habit (and control): a multi-level learning perspective on the
modulation of congruency effects. Frontiers in Psychology, 5.
https://doi.org/10.3389/fpsyg.2014.01247
Elsner, B., & Hommel, B. (2001). Effect anticipation and action control. Journal of
Experimental Psychology: Human Perception and Performance, 27(1), 229240.
https://doi.org/10.1037/0096-1523.27.1.229
Elsner, B., Hommel, B., Mentschel, C., Drzezga, A., Prinz, W., Conrad, B., & Siebner, H.
(2002). Linking Actions and Their Perceivable Consequences in the Human
Brain. NeuroImage, 17(1), 364372. https://doi.org/10.1006/nimg.2002.1162
Elsner, B., & Hommel, B. (2004). Contiguity and contingency in action-effect
learning. Psychological Research, 68(23), 138154.
https://doi.org/10.1007/s00426-003-0151-8
Espy, K. A. (2004). Using Developmental, Cognitive, and Neuroscience Approaches to
Understand Executive Control in Young Children. Developmental
Neuropsychology, 26(1), 379384. https://doi.org/10.1207/s15326942dn2601_1
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical
power analysis program for the social, behavioral, and biomedical
sciences. Behavior Research Methods, 39, 175-191.
One-shot learning of proactive effect monitoring 29
Findlay, J. M., & Walker, R. (1999). A model of saccade generation based on parallel
processing and competitive inhibition. Behavioral and Brain Sciences, 22(4), 661
674. https://doi.org/10.1017/s0140525x99002150
Gouret, F. & Pfeuffer, C. U. (2021). Learning to Expect the Future: Linking learning,
monitoring, and action control by assessing how fast anticipatory saccades
towards future action consequences emerge. Journal of Experimental Psychology:
Human Perception and Performance, preprint available online on ResearchGate.
https://doi.org/10.17605/OSF.IO/457GY
Haering, C., & Kiesel, A. (2014). Intentional Binding is independent of the validity of the
action effect’s identity. Acta Psychologica, 152, 109119.
https://doi.org/10.1016/j.actpsy.2014.07.015
Halekoh, U., & Højsgaard, S. (2014). A kenward-roger approximation and parametric
bootstrap methods for tests in linear mixed modelsthe R package
pbkrtest. Journal of Statistical Software, 59(9), 1-30.
Hayashi, M. J., Kantele, M., Walsh, V., Carlson, S., & Kanai, R. (2014). Dissociable
Neuroanatomical Correlates of Subsecond and Suprasecond time
Perception. Journal of Cognitive Neuroscience, 26(8), 16851693.
https://doi.org/10.1162/jocn_a_00580
Hayhoe, M. M., Shrivastava, A., Mruczek, R., & Pelz, J. B. (2003). Visual memory and motor
planning in a natural task. Journal of Vision, 3(1), 6. https://doi.org/10.1167/3.1.6
Herwig, A., & Horstmann, G. (2011). Actioneffect associations revealed by eye
movements. Psychonomic Bulletin & Review, 18(3), 531537.
https://doi.org/10.3758/s13423-011-0063-3
One-shot learning of proactive effect monitoring 30
Hommel, B., Müsseler, J., Aschersleben, G., & Prinz, W. (2001). The Theory of Event
Coding (TEC): A framework for perception and action planning. Behavioral And
Brain Sciences, 24(5), 849-878. https:/doi.org/10.1017/s0140525x01000103
Hommel, B. (2009). Action control according to TEC (theory of event
coding). Psychological Research, 73(4), 512-526. https:/doi.org/10.1007/s00426-
009-0234-2
Hommel, B. (2015). Between Persistence and Flexibility: The Yin and Yang of Action
Control (pp.33-67) in A. J. Elliot (Ed.), Advances in Motivation Science. Elsevier.
https://doi.org/10.1016/bs.adms.2015.04.003
Huestegge, L., & Kreutzfeldt, M. (2012). Action effects in saccade control. Psychonomic
Bulletin & Review, 19(2), 198203. https://doi.org/10.3758/s13423-011-0215-5
Janczyk, M., Durst, M., & Ulrich, R. (2016). Action selection by temporally distal goal
states. Psychonomic Bulletin & Review, 24(2), 467473.
https://doi.org/10.3758/s13423-016-1096-4
Jiang, J., Heller, K., & Egner, T. (2014). Bayesian modeling of flexible cognitive
control. Neuroscience & Biobehavioral Reviews, 46, 3043.
https://doi.org/10.1016/j.neubiorev.2014.06.001
Kenward, M. G., & Roger, J. H. (1997). Small Sample Inference for Fixed effects from
Restricted Maximum Likelihood. Biometrics, 53(3), 983.
https://doi.org/10.2307/2533558
Krigolson, O. E., & Holroyd, C. B. (2007). Hierarchical error processing: Different errors,
different systems. Brain Research, 1155, 70-80.
https://doi.org/10.1016/j.brainres.2007.04.024
One-shot learning of proactive effect monitoring 31
Kunde, W. (2001). Response-effect compatibility in manual choice reaction tasks. Journal of
Experimental Psychology: Human Perception and Performance, 27(2), 387394.
https://doi.org/10.1037/0096-1523.27.2.387
Kunde, W. (2003). Sequential modulations of stimulus-response correspondence effects
depend on awareness of response conflict. Psychonomic Bulletin & Review, 10(1),
198205. https://doi.org/10.3758/bf03196485
Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). lmerTest Package: Tests in
Linear Mixed effects Models. Journal of Statistical Software, 82(13).
https://doi.org/10.18637/jss.v082.i13
Land, M. F., & Hayhoe, M. (2001). In what ways do eye movements contribute to everyday
activities? Vision Research, 41(25-26), 35593565. https://doi.org/10.1016/s0042-
6989(01)00102-x
Land, M., Mennie, N., & Rusted, J. (1999). The Roles of Vision and Eye Movements in the
Control of Activities of Daily Living. Perception, 28(11), 13111328.
https://doi.org/10.1068/p2935
Lenth, R., & Lenth, M. R. (2018). Package ‘lsmeans’. The American Statistician, 34(4), 216-
221.
Miller, E. K., & Cohen, J. D. (2001). An Integrative Theory of Prefrontal Cortex
Function. Annual Review of Neuroscience, 24(1), 167202.
https://doi.org/10.1146/annurev.neuro.24.1.167
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D.
(2000). The Unity and Diversity of Executive Functions and Their Contributions to
Complex “Frontal Lobe” Tasks: A Latent Variable Analysis. Cognitive
Psychology, 41(1), 49100. https://doi.org/10.1006/cogp.1999.0734
One-shot learning of proactive effect monitoring 32
Moore, J., & Haggard, P. (2008). Awareness of action: Inference and
prediction. Consciousness and Cognition, 17(1), 136144.
https://doi.org/10.1016/j.concog.2006.12.004
Norman, D. A., & Shallice, T. (1986). Attention to Action. Consciousness and Self-
Regulation, 1–18. https://doi.org/10.1007/978-1-4757-0629-1_1
Nakagawa, S., Johnson, P. C. D., & Schielzeth, H. (2017). The coefficient of determination R
2 and intra-class correlation coefficient from generalized linear mixed-effects models
revisited and expanded. Journal of The Royal Society Interface, 14(134), 20170213.
https://doi.org/10.1098/rsif.2017.0213
Nakagawa, S., & Schielzeth, H. (2012). A general and simple method for obtainingR2from
generalized linear mixed-effects models. Methods in Ecology and Evolution, 4(2),
133142. https://doi.org/10.1111/j.2041-210x.2012.00261.x
Pfeuffer, C. U., Aufschnaiter, S., Thomaschke, R., & Kiesel, A. (2020). Only time will tell the
future: Anticipatory saccades reveal the temporal dynamics of time-based location
and task expectancy. Journal of Experimental Psychology: Human Perception and
Performance, 46(10), 11831200. https://doi.org/10.1037/xhp0000850
Pfeuffer, C. U., Huestegge, L., & Kiesel, A. (in preparation/2019). Deriving the Future from
the Past to Control the Present: The dynamic interplay of action selection and effect
monitoring in action control. Manuscript in preparation. Prior conference
presentation : On the proactive monitoring of future action consequences and its
coupling with effect-generating actions. Conference of the European Society for
Cognitive Psychology 2019, 21, 118. Retrieved from
http://escop2019.webs.ull.es/wp-content/uploads/2019/09/ESCoP2019-
ABSTRACTS.pdf
Pfeuffer, C. U., Kiesel, A., & Huestegge, L. (2016). A look into the future: Spontaneous
anticipatory saccades reflect processes of anticipatory action control. Journal of
One-shot learning of proactive effect monitoring 33
Experimental Psychology: General, 145(11), 15301547.
https://doi.org/10.1037/xge0000224
Pfeuffer, C. U., Kiesel, A., & Huestegge, L. (under revision/preprint). Proactive effect
Monitoring does not Differ between Free Choice and Forced Choice Action Modes.
Manuscript under review, preprint available online on ResearchGate.
https://doi.org/10.13140/RG.2.2.19875.48167/1
Pfeuffer, C. U., Moutsopoulou, K., Waszak, F., & Kiesel, A. (2019). Execution-based and
verbal code-based stimulusresponse associations: proportion manipulations
reveal conflict adaptation processes in item-specific priming. Psychological
Research, 84(8), 21722195. https://doi.org/10.1007/s00426-019-01220-3
Riechelmann, E., Pieczykolan, A., Horstmann, G., Herwig, A., & Huestegge, L. (2017).
Spatio-temporal dynamics of action-effect associations in oculomotor control. Acta
Psychologica, 180, 130136. https://doi.org/10.1016/j.actpsy.2017.09.003
Shin, Y. K., & Proctor, R. W. (2012). Testing boundary conditions of the ideomotor
hypothesis using a delayed response task. Acta Psychologica, 141(3), 360372.
https://doi.org/10.1016/j.actpsy.2012.09.008
Singmann, H., Bolker, B., Westfall, J., & Aust, F. (2015). afex: Analysis of factorial
experiments. R package version 0.13145.
Tatler, B. W., Brockmole, J. R., & Carpenter, R. H. S. (2017). LATEST: A model of saccadic
decisions in space and time. Psychological Review, 124(3), 267300.
https://doi.org/10.1037/rev0000054
Thomaschke, R., & Dreisbach, G. (2015). The time-event correlation effect is due to
temporal expectancy, not to partial transition costs. Journal of Experimental
Psychology: Human Perception and Performance, 41(1), 196218.
https://doi.org/10.1037/a0038328
One-shot learning of proactive effect monitoring 34
van Ede, F., Chekroud, S. R., Stokes, M. G., & Nobre, A. C. (2018). Decoding the influence
of anticipatory states on visual perception in the presence of temporal distractors.
Nature Communications, 9(1). https://doi.org/10.1038/s41467-018-03960-z
Vanneste, S., Baudouin, A., Bouazzaoui, B., & Taconnat, L. (2015). Age-related differences
in time-based prospective memory: The role of time estimation in the clock
monitoring strategy. Memory, 24(6), 812825.
https://doi.org/10.1080/09658211.2015.1054837
Verguts, T., & Notebaert, W. (2008). Hebbian learning of cognitive control: Dealing with
specific and nonspecific adaptation. Psychological Review, 115(2), 518525.
https://doi.org/10.1037/0033-295x.115.2.518
Waszak, F., & Herwig, A. (2007). Effect anticipation modulates deviance processing in the
brain. Brain Research, 1183, 7482. https://doi.org/10.1016/j.brainres.2007.08.082
Whitehead, P. S., Pfeuffer, C. U., & Egner, T. (2020). Memories of control: One-shot
episodic learning of item-specific stimulus-control associations. Cognition, 199,
104220. https://doi.org/10.1016/j.cognition.2020.104220
Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer.
Wirth, R., Pfister, R., Janczyk, M., & Kunde, W. (2015). Through the portal: effect
anticipation in the central bottleneck. Acta Psychologica, 160, 141151.
https://doi.org/10.1016/j.actpsy.2015.07.007
Wolfensteller, U., & Ruge, H. (2011). On the timescale of stimulus-based actioneffect
learning. Quarterly Journal of Experimental Psychology, 64(7), 12731289.
https://doi.org/10.1080/17470218.2010.546417
Yeung, N., Botvinick, M. M., & Cohen, J. D. (2004). The Neural Basis of Error Detection:
Conflict Monitoring and the Error-Related Negativity. Psychological Review, 111(4),
931959. https://doi.org/10.1037/0033-295x.111.4.931
One-shot learning of proactive effect monitoring 35
Figure 1. Trial structure: A forced-choice repeat/switch target was followed by a blank screen
response frame. Participants’ correct left/right responses produced visual effects on the
left/right side after a short (200 ms) or long (800 ms) effect delay. Effects appeared spatially
compatible with responses (e.g., left response -> effect on the left side) and each response
was mapped to one effect delay (e.g., left response -> 200 ms, right response -> 800 ms).
Trials were separated by a jittered intertrial interval (ITI). The mapping between participants’
responses and their respective effect delays switched after a sequence of either 4, 8, or 12
trials (sequences were randomly intermixed).
One-shot learning of proactive effect monitoring 36
Figure 2: Saccade-effect congruency (SEC) effects, that is, the percentage of saccades
towards the effect’s future location as a function of A) response occurrence and effect delay
One-shot learning of proactive effect monitoring 37
(200 ms vs. 800 ms) and B) participants’ time estimation scores per response occurrence
and effect delay. Response occurrence corresponds to the Xth time a response was correctly
performed in a sequence leading to the respective effect; 1-6). For instance, the second time
participants correctly performed a right response in a sequence was the second response
occurrence. A value of 50% (grey dashed line) indicates that an equal number of saccades
were directed towards and away from the future effect, as would be expected by chance.
Values above 50% indicate that participants looked more often towards future effects than
away from them, that is, participants anticipatorily saccaded towards future effects. Low time
estimation scores indicate temporal underestimations, whereas high scores indicate
accurate time estimations to temporal overestimations. A) Error bars and B) error bands
depict the 95% confidence interval. Violin shapes in A) depict the distribution of effect-
congruent saccades per condition.
One-shot learning of proactive effect monitoring 38
Figure 3: Mean latency of the first effect-congruent saccade as a function of A) response
occurrence and effect delay (200 ms vs. 800 ms) and B) participants’ time estimation scores
One-shot learning of proactive effect monitoring 39
per response occurrence, and effect delay. Response occurrence corresponds to the Xth
time a response was correctly performed in a sequence leading to the respective effect; 1-
6). For instance, the second time participants correctly performed a right response in a
sequence was the second response occurrence. Low time estimation scores indicate
temporal underestimations, whereas high scores indicate accurate time estimations to
temporal overestimations. A) Error bars and B) error bands depict the 95% confidence
interval. Violin shapes in A) depict the distribution of saccade latencies per condition.
One-shot learning of proactive effect monitoring 40
Appendix A – Model Specifications
Analyses were performed using Visual Code version 1.55.2 for R version 4.0.5 (R
Core Team, 2020). Linear mixed models (LMMs) were used to assess manual reaction time
and saccade latency and generalized linear mixed model (GLMMs) were used to assess
errors and saccade-effect congruency. (G)LMMs were computed with the lme4 (Bates et al.,
2015), lmerTest (Kuznetsova et al., 2017), pbkrtest (Halekoh & Højsgaard, 2014), afex
(Singmann et al., 2015), and emmeans (Lenth et al., 2018) packages. To create figures, we
used the ggplot2 package (Wickham, 2016). For LMMs, we used the maximum likelihood
estimation and the Satterthwaite (Kuznetsova et al., 2017) method to assess p-values for
model selection and restricted maximum likelihood estimation and the Kenward-Roger
(Kenward & Roger, 1997) approximation for denominator degrees of freedom to examine the
final model. For GLMMs, we used maximum likelihood estimation and binomial link functions
(bobyqa optimizer, 1,000,000 iterations) and p-values were estimated via asymptotic Wald
tests. For the GLMM, we report odds ratios (Szumilas, 2010). Conditional R2 values for each
mixed model were computed using the r.squaredGLMM function of the MuMIn package
(Nakagawa & Schielzeth, 2013; see also Nakagawa et al., 2017).
Per dependent variable, we conducted a (G)LMM analysis featuring effect delay (200
ms vs. 800 ms) and response occurrence (corresponding to the Xth time participants
correctly performed the respective response in the respective sequence leading to an effect,
1-6) as well as their interaction as fixed effects. In addition, participants’ individual, z-
standardized time estimation score (see following section for computation details) and its
interactions with the other predictors were included as fixed effects. Only correct responses
that produced the respective effect were counted for this analysis, so that response
occurrence reflected the number of times that participants had already experienced the
response to produce the respective effect in that sequence (response occurrence = 1-6; e.g.,
response occurrence 1 = 0 prior experiences of the effect following this response in the
current sequence). The independent variables effect delay (200 ms = -1, 800 ms = 1) and
One-shot learning of proactive effect monitoring 41
response occurrence (response occurrence 1 = -2.5, 2 = -1.5, 3 = -0.5, 4 = 0.5, 5 = 1.5, 6 =
2.5) were contrast coded for these analyses. Effect coding was used for effect delay and
simple coding was used for response occurrence (reference condition: response occurrence
1). That is, for response occurrence, we coded a matrix of five contrasts comparing
response occurrence 1 with response occurrence 2-6, respectively, while taking the grand
mean into account (i.e., the intercept represents the grand mean of all conditions).
Per (G)LMM, we first implemented a random-effects structure including participant
intercepts and by-participant random slopes for effect delay and response occurrence (see
Barr et al., 2013). When the respective (G)LMM did not converge (i.e., without a singular fit
and without negative Hessian eigenvalues), we first removed correlations among random
slopes, then the by-participant random slopes of response occurrence, and finally the by-
participant random slopes of effect delay until the model converged without a singular fit or a
negative Hessian eigenvalue.
One-shot learning of proactive effect monitoring 42
Appendix B
Manual responses
Trials with premature (<0.1%) or omitted (0.7%) responses were excluded from all
analyses. For the analyses of manual RTs and anticipatory saccades, trials containing
errors (4.7%) were additionally excluded. Furthermore, trials with RTs deviating by more
than three SDs from their individual cell means were considered as outliers and excluded
from the RT analysis (1.7%).
Errors. For the analysis of errors only, response occurrence s included trials with
errors. That is, response occurrence corresponds to the Xth time a response was performed
(regardless of its correctness) in a sequence. The GLMM fitting manual errors (0 = correct
response, 1 = error) included all interactions of the fixed effect delay and response
occurrence and the continuous predictor time estimation ability as well as participant
intercepts as random effects (see Figure B1).
Effect delay and time estimation score significantly interacted, z = 2.84, p = .005, OR
= 1.13. Time estimation slopes were steeper for long as compared to short effect delays.
That is, the more participants underestimated effect delays, the more likely they committed
an error during the long as compared to the short effect delay. This pattern reversed for
accurate estimations or temporal overestimations.
Error rate differences between response occurrence 1 and every other response
occurrence were significant with fewer errors for response occurrence 1, response
occurrence 1 vs. 2: z = 3.22, p = .001, OR = 1.58; response occurrence 1 vs. 3: z = 2.45, p =
.014, OR = 1.45; response occurrence 1 vs. 4: z = 2.33, p = .020, OR = 1.42; response
occurrence 1 vs. 5: z = 2.61, p = .009, OR = 1.58, response occurrence 1 vs. 6: z = 2.98, p =
.003, OR = 1.67. That is, participants performed fewer errors at the beginning of a sequence
(response occurrence 1 occurred very early in the sequence in most cases) when they had
One-shot learning of proactive effect monitoring 43
just experienced the switch in action-effect delay mapping as compared to later on in the
sequence.
The interaction of time estimation and response occurrence 1 vs. 3, z = -2.72, p =
.007, OR = 0.66, as well as the interaction of time estimation and response occurrence 1 vs.
5, z = -2.72, p = .006, OR = 0.62, were significant. For response occurrence 1, the time
estimation slope was positive, whereas it was negative for response occurrences 3 and 5.
None of the other effects or their interactions reached significance, zs 1.86, ps
.063 (see Table 3 in Supplementary Materials).
Reaction times (RTs). The LMM fitting manual RTs included all interactions of the
fixed effects effect delay and response occurrence and the continuous predictor time
estimation ability as well as participant intercepts as random effects (see Figure B2). Manual
RTs significantly differed between response occurrence 1 and every other response
occurrence , response occurrence 1 vs. 2: t = 4.79, p < .001, = 31 ms; response
occurrence 1 vs. 3: t = 3.99, p < .001, = 29 ms; response occurrence 1 vs. 4: t = 2.55, p =
.011, = 19 ms; response occurrence 1 vs. 5: t = 2.75, p = .006, = 24 ms, response
occurrence 1 vs. 6: t = 1.98, p = .048, = 18 ms. That is, participants responded faster at
the beginning of a sequence (response occurrence 1 occurred very early in the sequence in
most cases) when they had just experienced the switch in action-effect delay mapping as
compared to later on in the sequence.
The interactions of effect delay and the comparison of response occurrence 1 with
the other response occurrences were significant for response occurrence 1 vs. 2, t = -2.11, p
= .035, = 14 ms, for response occurrence 1 vs. 4, t = -2.55, p = .011, = 19 ms, and for
response occurrence 1 vs. 5, t = -2.47, p = .014, = 21 ms, other response occurrences: ts
0.30, ps.390. Post-hoc tests showed that RTs significantly differed between short and
long effect delays for response occurrence 1, t = 2.14, p = .032, = 12 ms, but not for
One-shot learning of proactive effect monitoring 44
response occurrence 2, t = -0.64, p = .519, = 3 ms, response occurrence 4, t = -1.44, p =
.150, = 7 ms, and response occurrence 5, t = -1.44, p = .149, = -10 ms.
None of the other effects or their interactions reached significance, ts 1.43, ps
.154 (see Table 4 in Supplementary Materials)).
Discussion.
Considering that manual responses reflect influences of action-effect associations on
action selection (of the effect-generating, here manual, action; e.g., Hommel & Elsner, 2001;
Hommel et al., 2001; Kunde, 2001), our results showed that manual RTs were not
systematically affected by effect delay individually, across response occurrences, or based
on individual time estimation abilities response occurrence time estimation. This finding is in
line with prior studies on action-effect learning (assessing the learning of effect location)
which suggested that at minimum eight pairings of action and effect were necessary to
observe effects on action selection (Wolfensteller & Ruge, 2011; see also Gouret & Pfeuffer,
2021).
This is the first study to assess the emergence of action-effect delay associations in
the range of a few trials. For manual RTs, our results parallel findings for action-effect
location associations, suggesting that our at most six action-effect delay mappings were
insufficient to yield systematic influences on action selection. However, in error rates, we did
find a non-significant trend of effect delay and a significant interaction between effect delay
and participants’ time estimations. These results suggests that action-effect delay
associations might be able to affect action selection (at least response accuracy) earlier than
action-effect location associations. Future studies including somewhat longer sequences will
need to replicate this finding and extend on it to confirm this assumption.
One-shot learning of proactive effect monitoring 45
One-shot learning of proactive effect monitoring 46
Figure B1. Error rates in manual response as A) a function of response occurrence and
effect delay (200 ms vs. 800 ms) and B) participants’ time estimation scores per response
occurrence, and effect delay. Response occurrence corresponds to the Xth time a response
was correctly performed in a sequence leading to the respective effect; 1-6). For instance,
the second time participants correctly performed a right response in a sequence was the
second response occurrence. Low time estimation scores indicate temporal
underestimations, whereas high scores indicate accurate time estimations to temporal
overestimations. A) Error bars and B) error bands depict the 95% confidence interval. Violin
shapes in A) depict the distribution of errors per condition.
One-shot learning of proactive effect monitoring 47
Figure B2: Manual reaction times (RTs) in manual response as A) a function of response
occurrence and effect delay (200 ms vs. 800 ms) and B) participants’ time estimation scores
One-shot learning of proactive effect monitoring 48
per response occurrence, and effect delay. Response occurrence corresponds to the Xth
time a response was correctly performed in a sequence leading to the respective effect; 1-
6). For instance, the second time participants correctly performed a right response in a
sequence was the second response occurrence. A) Error bars and B) error bands depict the
95% confidence interval. Violin shapes in A) depict the distribution of manual reaction times
per condition.
One-shot learning of proactive effect monitoring 49
Appendix C
Saccade-manual latency differences
According to Pfeuffer et al. (2016, see also Gouret & Pfeuffer, 2021, Pfeuffer et al.,
under revision/preprint), saccade latency and manual RT are strongly correlated and
anticipatory saccades are on average performed after manual response execution. This led
Pfeuffer et al. to conclude that saccade latencies might be affected by effects on manual
RTs that propagate. Thus, we additionally assessed saccade-manual latency differences
between saccade latencies and manual RTs on corresponding trials latency = latency1st
effect−congruent saccadeRTmanual) to rule out a possible influence.
The LMM on Δ latency of participants’ first effect-congruent saccade per trial included
all interactions of the fixed effect delay and response occurrence and the continuous
predictor time estimation ability as well as both participant intercepts and by-participant
random slopes for effect delay as random effects (see Figure C1).
Δ latency was significantly longer for long (800 ms) as compared to short (200 ms)
effect delays, t = -4.65, p < .001, = 109 ms. Moreover, Δ latency differences between the
short and long effect delay were more pronounced the more participants tended to
underestimate effect delays, t = 2.48, p = .021, = 58 ms. In a post-hoc test, this was
indicated by steeper slopes for long effect delays as compared to short effect delays, t =
2.49, p = .015.
No other main effects or their interactions reached significance, ts 1.54, ps ≥ .070
(see Table 5 in Supplementary Materials for model details).
Discussion.
In contrast to prior studies (Gouret & Pfeuffer, 2021; Pfeuffer et al., 2016; under
revision/preprint, in preparation/2019) where participants consistently performed eye
movements during late stages of motor preparation (i.e., around the time of or after manual
responses), in the present experiment, participants, on average performed anticipatory
One-shot learning of proactive effect monitoring 50
saccades significantly prior to manual responses (effect delay: 200 ms) or around the time of
manual responses (effect delay: 800 ms). In prior studies, this was neither the case when
effect delay was manipulated (Pfeuffer et al., in preparation/2019) nor when action-effect
location mappings frequently switched (Gouret & Pfeuffer, 2021). It might therefore point
towards a strong influence of the volatility of action-effect delay mappings on the temporal
scheduling of anticipatory saccades relative to manual responses. Yet, further experiments
will be necessary to confirm this assumption.
One-shot learning of proactive effect monitoring 51
Figure C1: Mean latency of the first effect-congruent saccade (
) as a function of A) response occurrence and
effect delay (200 ms vs. 800 ms) and B) participants’ time estimation scores per response
One-shot learning of proactive effect monitoring 52
occurrence and effect delay. Response occurrence corresponds to the Xth time a response
was correctly performed in a sequence leading to the respective effect; 1-6). For instance,
the second time participants correctly performed a right response in a sequence was the
second response occurrence. Low time estimation scores indicate temporal
underestimations, whereas high scores indicate accurate time estimations to temporal
overestimations. A) Error bars and B) error bands depict the 95% confidence interval. Violin
shapes in A) depict the distribution of Δ latency per condition.
One-shot learning of proactive effect monitoring 53
Supplementary material
Figures
Figure 1: Violin plot depicting the frequency distribution of saccade end positions (saccadic
gain: 0 = screen center, 1 = effect position, 1 = opposite position) per effect-delay condition
(short vs. long) and response occurrence. Response occurrence corresponds to the Xth time
a response was correctly performed in a sequence leading to the respective effect; 1-6). For
instance, the second time participants correctly performed a right response in a sequence
was the second response occurrence. Saccade end positions were mapped, so that positive
saccadic gain values indicate saccades toward the effect (green/upper dashed line = effect
position) and negative values indicate saccades away from the effect (orange/lower dashed
line = opposite position). See the online article for the colour version of this figure.
One-shot learning of proactive effect monitoring 54
Figure 2: Mean latency of the first effect-congruent saccade as a function of trial position within a
trial sequence (trial 2 - 12) and effect delay (200 ms vs. 800 ms). Error bars depict the 95%
confidence interval of saccade latencies per position in the sequence (sequence length of 4, 8, or 12
trials; i.e., trials 2-4/8/12). The first trial of each sequence is excluded as participants are not yet
aware of the change in action-effect delay mappings before they have experienced the effect of
their response at the end of sequence trial 1.
One-shot learning of proactive effect monitoring 55
Mixed Model Tables
Table 1
Saccade-Effect Congruency (SEC; 0/1)
Predictors
OR
CI
SE
z
p
Intercept
9.29
4.9317.51
0.32
6.90
<0.001
response occurrence 1 vs. 2
1.04
0.821.31
0.12
0.33
0.741
response occurrence 1 vs. 3
1.16
0.901.50
0.13
1.13
0.258
response occurrence 1 vs. 4
0.83
0.651.07
0.13
-1.43
0.152
response occurrence 1 vs. 5
0.79
0.591.04
0.14
-1.66
0.097
response occurrence 1 vs. 6
1.04
0.751.45
0.17
0.25
0.803
effect delay
1.04
0.961.13
0.04
0.95
0.344
time estimation
1.85
0.983.48
0.32
1.90
0.057
response occurrence 1 vs. 2
x effect delay
1.08
0.851.36
0.12
0.64
0.522
response occurrence 1 vs. 3
x effect delay
0.98
0.751.27
0.13
-0.17
0.864
response occurrence 1 vs. 4
x effect delay
1.01
0.791.29
0.13
0.08
0.933
response occurrence 1 vs. 5
x effect delay
1.08
0.811.43
0.14
0.52
0.603
response occurrence 1 vs. 6
x effect delay
1.16
0.831.61
0.17
0.88
0.379
response occurrence 1 vs. 2
x time estimation
1.05
0.831.33
0.12
0.43
0.668
response occurrence 1 vs. 3
x time estimation
1.12
0.861.45
0.13
0.86
0.390
response occurrence 1 vs. 4
x time estimation
0.96
0.751.23
0.13
-0.29
0.771
response occurrence 1 vs. 5
x time estimation
0.96
0.721.28
0.15
-0.28
0.780
response occurrence 1 vs. 6
x time estimation
1.12
0.801.55
0.17
0.65
0.513
One-shot learning of proactive effect monitoring 56
effect delay x time estimation
0.96
0.891.04
0.04
-0.92
0.360
response occurrence 1 vs. 2
x effect delay x time
estimation
1.08
0.851.36
0.12
0.63
0.532
response occurrence 1 vs. 3
x effect delay x time
estimation
0.95
0.731.23
0.13
-0.39
0.694
response occurrence 1 vs. 4
x effect delay x time
estimation
0.89
0.691.13
0.13
-0.96
0.338
response occurrence 1 vs. 5
x effect delay x time
estimation
1.09
0.821.45
0.14
0.60
0.546
response occurrence 1 vs. 6
x effect delay x time
estimation
0.98
0.711.36
0.17
-0.12
0.908
Model
σ2
3.29
τ00subject
2.54
N subject
25
Observations
10082
Marginal R2 / Conditional R2
0.064 / 0.472
Deviance
5630.717
Note. CI indicates confidence interval and SE refers to standard error.
Final model: SEC ~ response occurrence * effect delay * time estimation + (1 | Subject)
One-shot learning of proactive effect monitoring 57
Table 2
Saccade Latency
Predictors
Estimate
CI
SE
t
p
Intercept
544.74
503.09586.38
20.14
27.05
<0.001
response occurrence 1 vs. 2
13.23
0.7625.71
6.36
2.08
0.038
response occurrence 1 vs. 3
23.29
9.9036.69
6.83
3.41
0.001
response occurrence 1 vs. 4
8.70
-4.9622.35
6.97
1.25
0.212
response occurrence 1 vs. 5
9.10
-6.8725.07
8.14
1.12
0.264
response occurrence 1 vs. 6
3.10
-14.0920.29
8.77
0.35
0.724
effect delay
-49.16
-69.83 – -28.50
10.00
-4.92
<0.001
time estimation
-36.47
-78.195.25
20.18
-1.81
0.084
response occurrence 1 vs. 2
x effect delay
-4.51
-16.987.96
6.36
-0.71
0.479
response occurrence 1 vs. 3
x effect delay
-10.88
-24.282.51
6.83
-1.59
0.111
response occurrence 1 vs. 4
x effect delay
-10.46
-24.123.19
6.97
-1.50
0.133
response occurrence 1 vs. 5
x effect delay
-15.38
-31.340.59
8.14
-1.89
0.059
response occurrence 1 vs. 6
x effect delay
-3.13
-20.3214.05
8.77
-0.36
0.721
response occurrence 1 vs. 2
x time estimation
-4.54
-17.418.33
6.57
-0.69
0.489
response occurrence 1 vs. 3
x time estimation
-10.16
-24.003.69
7.06
-1.44
0.151
response occurrence 1 vs. 4
x time estimation
-1.86
-15.9212.20
7.17
-0.26
0.795
response occurrence 1 vs. 5
x time estimation
-7.84
-24.498.81
8.49
-0.92
0.356
response occurrence 1 vs. 6
x time estimation
1.72
-15.9219.36
9.00
0.19
0.849
effect delay x time estimation
23.00
2.2243.78
10.06
2.29
0.032
One-shot learning of proactive effect monitoring 58
response occurrence 1 vs. 2
x effect delay x time
estimation
11.14
-1.7324.01
6.57
1.70
0.090
response occurrence 1 vs. 3
x effect delay x time
estimation
8.45
-5.4022.30
7.06
1.20
0.232
response occurrence 1 vs. 4
x effect delay x time
estimation
1.27
-12.7815.33
7.17
0.18
0.859
response occurrence 1 vs. 5
x effect delay x time
estimation
5.31
-11.3421.96
8.49
0.63
0.532
response occurrence 1 vs. 6
x effect delay x time
estimation
-3.31
-20.9514.33
9.00
-0.37
0.713
Model
σ2
32020.20
τ00subject
9931.06
N subject
25
Observations
8609
Marginal R2 / Conditional R2
0.087 / 0.304
Deviance
113904.446
Note. CI indicates confidence interval and SE refers to standard error.
Final model: Saccade latency ~ response occurrence * effect delay * time estimation +
(effect delay||subject)
One-shot learning of proactive effect monitoring 59
Table 3
Manual Response – errors (0/1)
Predictors
OR
CI
SE
z
p
Intercept
0.04
0.030.05
0.16
-21.34
<0.001
response occurrence 1 vs. 2
1.58
1.202.08
0.14
3.22
0.001
response occurrence 1 vs. 3
1.45
1.081.95
0.15
2.45
0.014
response occurrence 1 vs. 4
1.42
1.061.92
0.15
2.33
0.020
response occurrence 1 vs. 5
1.58
1.122.22
0.17
2.61
0.009
response occurrence 1 vs. 6
1.67
1.192.33
0.17
2.98
0.003
effect delay
0.93
0.851.01
0.04
-1.78
0.075
time estimation
0.94
0.701.28
0.15
-0.37
0.709
response occurrence 1 vs. 2
x effect delay
1.21
0.911.59
0.14
1.33
0.185
response occurrence 1 vs. 3
x effect delay
1.26
0.941.70
0.15
1.52
0.128
response occurrence 1 vs. 4
x effect delay
1.20
0.891.62
0.15
1.22
0.224
response occurrence 1 vs. 5
x effect delay
1.38
0.981.94
0.17
1.86
0.063
response occurrence 1 vs. 6
x effect delay
1.11
0.791.55
0.17
0.61
0.543
response occurrence 1 vs. 2
x time estimation
0.80
0.611.05
0.14
-1.60
0.110
response occurrence 1 vs. 3
x time estimation
0.66
0.490.89
0.15
-2.72
0.007
response occurrence 1 vs. 4
x time estimation
0.76
0.571.02
0.15
-1.83
0.067
response occurrence 1 vs. 5
x time estimation
0.62
0.440.87
0.18
-2.72
0.006
response occurrence 1 vs. 6
x time estimation
0.73
0.521.02
0.17
-1.86
0.063
effect delay x time estimation
1.13
1.041.23
0.04
2.84
0.005
One-shot learning of proactive effect monitoring 60
response occurrence 1 vs. 2
x effect delay x time
estimation
1.06
0.801.39
0.14
0.40
0.691
response occurrence 1 vs. 3
x effect delay x time
estimation
1.04
0.771.40
0.15
0.26
0.791
response occurrence 1 vs. 4
x effect delay x time
estimation
0.90
0.671.21
0.15
-0.69
0.489
response occurrence 1 vs. 5
x effect delay x time
estimation
1.27
0.901.79
0.18
1.35
0.177
response occurrence 1 vs. 6
x effect delay x time
estimation
0.87
0.621.22
0.17
-0.82
0.413
Model
σ2
3.29
τ00subject
0.49
N subject
25
Observations
16343
Marginal R2 / Conditional R2
0.021 / 0.160
Deviance
5675.696
Note. OR refers to odds ratio, CI indicates confidence interval, and
SE refers to standard error.
Final model: Error ~ response occurrence * effect delay * time estimation + (1|subject)
One-shot learning of proactive effect monitoring 61
Table 4
Manual ResponsesReaction Time
Predictors
Estimate
CI
SE
t
p
Intercept
579.23
547.16611.3
0
16.36
35.40
<0.001
response occurrence 1 vs. 2
15.94
9.4222.45
3.33
4.79
<0.001
response occurrence 1 vs. 3
14.30
7.2721.33
3.59
3.99
<0.001
response occurrence 1 vs. 4
9.28
2.1616.40
3.63
2.55
0.011
response occurrence 1 vs. 5
11.82
3.3920.25
4.30
2.75
0.006
response occurrence 1 vs. 6
9.11
0.0718.15
4.61
1.98
0.048
effect delay
0.54
-1.722.80
1.15
0.47
0.640
time estimation
2.78
-29.2934.86
16.37
0.17
0.866
response occurrence 1 vs. 2
x effect delay
-7.03
-13.55 – -0.51
3.33
-2.11
0.035
response occurrence 1 vs. 3
x effect delay
-3.09
-10.123.94
3.59
-0.86
0.390
response occurrence 1 vs. 4
x effect delay
-9.27
-16.39 – -2.15
3.63
-2.55
0.011
response occurrence 1 vs. 5
x effect delay
-10.60
-19.03 – -2.17
4.30
-2.47
0.014
response occurrence 1 vs. 6
x effect delay
-1.40
-10.427.63
4.61
-0.30
0.762
response occurrence 1 vs. 2
x time estimation
-1.52
-8.055.01
3.33
-0.46
0.648
response occurrence 1 vs. 3
x time estimation
-3.14
-10.183.90
3.59
-0.87
0.382
response occurrence 1 vs. 4
x time estimation
-0.59
-7.726.53
3.63
-0.16
0.870
response occurrence 1 vs. 5
x time estimation
-6.16
-14.622.31
4.32
-1.43
0.154
response occurrence 1 vs. 6
x time estimation
-1.42
-10.417.56
4.58
-0.31
0.756
effect delay x time estimation
-1.18
-3.441.07
1.15
-1.03
0.304
One-shot learning of proactive effect monitoring 62
response occurrence 1 vs. 2
x effect delay x time
estimation
-0.74
-7.285.79
3.33
-0.22
0.824
response occurrence 1 vs. 3
x effect delay x time
estimation
-1.32
-8.365.73
3.59
-0.37
0.714
response occurrence 1 vs. 4
x effect delay x time
estimation
2.84
-4.299.96
3.63
0.78
0.435
response occurrence 1 vs. 5
x effect delay x time
estimation
1.43
-7.039.90
4.32
0.33
0.740
response occurrence 1 vs. 6
x effect delay x time
estimation
-3.36
-12.345.62
4.58
-0.73
0.463
Model
σ2
16895.22
τ00subject
6659.45
N subject
25
Observations
15346
Marginal R2 / Conditional R2
0.002 / 0.284
Deviance
193053.338
Note. OR refers to odds ratio, CI indicates confidence interval, and
SE refers to standard error.
Final model: RT ~ response occurrence * effect delay * time estimation + (1|subject)
One-shot learning of proactive effect monitoring 63
Table 5
Saccade-manual Latency Differences Latency)
Predictors
Estimate
CI
SE
t
p
Intercept
-35.60
-75.314.10
19.20
-1.85
0.076
response occurrence 1 vs.
2
1.28
-9.3711.93
5.43
0.24
0.814
response occurrence 1 vs.
3
8.99
-2.4420.43
5.83
1.54
0.123
response occurrence 1 vs.
4
4.98
-6.6816.64
5.95
0.84
0.403
response occurrence 1 vs.
5
4.87
-8.7618.50
6.95
0.70
0.484
response occurrence 1 vs.
6
0.46
-14.2215.13
7.49
0.06
0.951
effect delay
-54.34
-78.53 – -30.15
11.70
-4.65
<0.001
time estimation
-36.53
-76.293.23
19.23
-1.90
0.070
response occurrence 1 vs.
2 x effect delay
3.71
-6.9414.36
5.43
0.68
0.495
response occurrence 1 vs.
3 x effect delay
-0.64
-12.0710.80
5.83
-0.11
0.913
response occurrence 1 vs.
4 x effect delay
1.78
-9.8813.44
5.95
0.30
0.765
response occurrence 1 vs.
5 x effect delay
-2.65
-16.2810.98
6.95
-0.38
0.703
response occurrence 1 vs.
6 x effect delay
1.21
-13.4715.89
7.49
0.16
0.872
response occurrence 1 vs.
2 x time estimation
-6.72
-17.714.27
5.61
-1.20
0.231
response occurrence 1 vs.
3 x time estimation
-9.22
-21.052.60
6.03
-1.53
0.126
response occurrence 1 vs.
4 x time estimation
-4.20
-16.207.81
6.12
-0.69
0.493
response occurrence 1 vs.
5 x time estimation
-10.64
-24.863.57
7.25
-1.47
0.142
One-shot learning of proactive effect monitoring 64
response occurrence 1 vs.
6 x time estimation
-9.44
-24.505.63
7.69
-1.23
0.220
effect delay x time
estimation
29.13
4.8653.39
11.74
2.48
0.021
response occurrence 1 vs.
2 x effect delay x time
estimation
1.84
-9.1512.83
5.61
0.33
0.743
response occurrence 1 vs.
3 x effect delay x time
estimation
3.41
-8.4215.23
6.03
0.57
0.572
response occurrence 1 vs.
4 x effect delay x time
estimation
-4.97
-16.987.03
6.12
-0.81
0.417
response occurrence 1 vs.
5 x effect delay x time
estimation
-0.06
-14.2814.16
7.25
-0.01
0.993
response occurrence 1 vs.
6 x effect delay x time
estimation
-4.35
-19.4210.71
7.69
-0.57
0.571
Model
σ2
23344.63
τ00Subject
9195.13
N Subject
25
Observations
8609
Marginal
R2 / Conditional R2
0.121 / 0.396
Deviance
111190.308
Note. CI indicates confidence interval and SE refers to standard error.
Final model: Δ latency ~ response occurrence * effect delay * time estimation + (Effect delay
|| Subject)
ResearchGate has not been able to resolve any citations for this publication.
Preprint
Full-text available
While performing an action that contingently yields the same effect, we form action-effect associations that allow us to anticipate the effects of our actions. Importantly, our eyes already move towards the expected future location of our actions´ effects in anticipation of them, that is, we perform anticipatory saccades. These anticipatory saccades are linked to a proactive effect monitoring process that prepares the comparison of expected and actual effect. However, how fast such anticipatory saccades emerge (i.e., how fast learning leads to monitoring) is unknown. To address this question, here, correct left/right responses were followed by a visual effect either on the same side (action-effect compatible) or on the opposite side (action-effect incompatible). In Experiment 1, action-effect compatibility switched after sequences of 4, 8, or 12 trials (randomly allocated; partly predictable environment). In Experiment 2, random trials (2 to 7) separated sequences of 3, 5, or 7 experimental trials. Again, action-effect compatibility switched after a sequence and sequences were randomly allocated (unpredictable environment). In both experiments, participants started to perform anticipatory saccades towards future effects after experiencing a new action-effect mapping just once to twice. Thus, one to two action-effect learning instances were sufficient to develop action-effect associations that trigger attentional shifts towards the expected future consequences of our actions (i.e., monitoring processes), whereas influences on action selection are only observed after a much larger number of learning instances. This pattern might suggest that monitoring processes modulate the expression of action-effect associations in action planning based on observed action-effect contingencies.
Preprint
Full-text available
When our actions yield predictable consequences in the environment, our eyes often already saccade towards the locations we expect these consequences to appear at. Such spontaneous anticipatory saccades occur based on bi-directional associations between action and effect formed by prior experience. That is, our eye movements are guided by expectations derived from prior learning history. Anticipatory saccades presumably reflect a proactive effect monitoring process that prepares a later comparison of expected and actual effect. Here, we examined whether anticipatory saccades emerged under forced choice conditions when only actions but not target stimuli were predictive of future effects and whether action mode (forced choice vs. free choice, i.e., stimulus-based vs. stimulus-independent choice) affected anticipatory saccades. Participants produced predictable visual effects on the left/right side via forced choice and free choice left/right key presses. Action and visual effect were spatially compatible in one half of the experiment and spatially incompatible in the other half. Irrespective of whether effects were predicted by target stimuli in addition to participants' actions, in both action modes, we observed anticipatory saccades towards the location of future effects. Importantly, neither the frequency, nor latency or amplitude of these anticipatory saccades significantly differed between forced choice and free choice action modes. Overall, our findings suggest that proactive effect monitoring of future action consequences as reflected in anticipatory saccades does not differ between forced choice and free choice action modes.
Article
Full-text available
Stimulus-response (S-R) associations consist of two independent components: Stimulus-classification (S-C) and stimulus-action (S-A) associations. Here, we examined whether these S-C and S-A associations were modulated by cognitive control operations. In two item-specific priming experiments, we systematically manipulated the proportion of trials in which item-specific S-C and/or S-A mappings repeated or switched between the single encoding (prime) and single retrieval (probe) instance of each stimulus (i.e., each stimulus appeared only twice). Thus, we assessed the influence of a list-level proportion switch manipulation on the strength of item-specific S-C and S-A associations. Participants responded slower and committed more errors when item-specific S-C or S-A mappings switched rather than repeated between prime and probe (i.e., S-C/S-A switch effects). S-C switch effects were larger when S-C repetitions rather than switches were frequent on the list-level. Similarly, S-A switch effects were modulated by S-A switch proportion. Most importantly, our findings rule out contingency learning and temporal learning as explanations of the observed results and point towards a conflict adaptation mechanism that selectively adapts the encoding and/or retrieval for each S-R component. Finally, we outline how cognitive control over S-R associations operates in the context of item-specific priming.
Article
Full-text available
Cognitive flexibility refers to the ability to quickly reconfigure our mind, like when we switch between different tasks. This review highlights recent evidence showing that cognitive flexibility can be conditioned by simple incentives typically known to drive lower-level learning, such as stimulus-response associations. Cognitive flexibility can also become associated with, and triggered by, bottom-up contextual cues in our environment, including subliminal cues. Therefore, we suggest that the control functions that mediate cognitive flexibility are grounded in, and guided by, basic associative learning mechanisms, and abide by the same learning principles as more low-level forms of behavior. Such a learning perspective on cognitive flexibility offers new directions and important implications for further research, theory, and applications.
Article
Full-text available
Adjustments in cognitive control, as measured by congruency sequence effects, are thought to be influenced by both external stimuli and internal goals. However, this dichotomy has oft