Skill learning can be independent of speed and accuracy instructions

Skill learning can be independent of speed and accuracy instructions Teodóra Vékony, Hanna Marossy, Anita Must, László Vécsei, Karolina Janacsek & Dezso Nemeth a Department of Neurology, University of Szeged, Szeged, Semmelweis utca 6, 6725 Szeged, Hungary b Institute of Psychology, ELTE Eötvös Loránd University, Izabella utca 46, 1064, Budapest, Hungary c Institute of Psychology, University of Szeged, Egyetem utca 2, 6722 Szeged, Hungary d MTA-SZTE Neuroscience Research Group, Semmelweis u. 6, H-6725 Szeged, Hungary e Brain, Memory and Language Research Group, Institute of Cognitive Neuroscience and Psychology, Research Centre for Natural Sciences, Hungarian Academy of Sciences, Budapest, Hungary


Introduction
Our social, motor and cognitive skills help us to adapt and function in various situations in our everyday life. Therefore, our ability to learn new skills is crucial, and fine-tuning this capability can be advantageous for an individual. Previous studies investigating sports performance (Beilock, Bertenthal, Hoerger, & Carr, 2008;Beilock, Bertenthal, McCoy, & Carr, 2004), implicit (Hoyndorf & Haider, 2009), and explicit sequence learning (Barnhoorn, Panzer, Godde, & Verwey, 2019) typically found that speed and accuracy strategies differently affects skill learning. However, skill learning is multifaceted, and it still not clear what underlying mechanisms benefit from speed and accuracy instructions and what mechanisms do not. A core component of learning new skills is picking up complex statistical regularities from the environment (Conway, 2020;Janacsek, Fiser, & Nemeth, 2012). So far, no study has investigated the effects of prioritizing speed or accuracy on the acquisition of such statistical dependencies. Do such instructions affect our performance or also the established statistical representations? Above the theoretical importance of answering this question, it also has clear practical implications. If instructions do affect statistical knowledge acquisition itself, then differences in the exact instruction between studies or the interpretation of instructions among participants might affect the conclusions about skill learning. Here, we aim to unveil how emphasizing speed and accuracy influence an essential aspect of skill learning, namely the acquisition of complex statistical representations. Hoyndorf and Haider (2009) investigated the sequencing aspect of skill learning and found accuracy strategy to impair the expression of implicit knowledge compared to speed instruction; however, learning was still detected under accuracy instruction compared to a non-learning control group. Yet in this experiment, the accumulated sequence-knowledge under speed/accuracy instructions was not compared to a phase where the importance of speed and accuracy was equally emphasized. Such a comparison would reveal whether 4 implicit sequence knowledge is acquired to the same level under different instructions.
Recently, Barnhoorn, Panzer, Godde, and Verwey (2019) found that speed instruction benefits the development of representations about repeating sequences while forcing participants to be more accurate leads to faster selection of responses via better stimulus-response associations.
In this study, the participants were aware of the repeating sequences; thus the learning was utterly explicit. Taken together, the studies mentioned above suggest that speed instruction might benefit sequence learning more than accuracy instruction. These studies used relatively simple, deterministic sequences (i.e., sequences with a simple repeating pattern). Therefore, data are still lacking on whether instruction affects more complex, probabilistic sequence representations.
Here, we aimed to test whether speed or accuracy instructions affect the acquisition of complex statistical regularities using an implicit probabilistic sequence learning task. We go beyond previous investigations by at least two aspects: First, by studying complex probabilistic sequences with non-adjacent, second-order dependencies (Remillard, 2008). This feature means that to predict the n th element of the sequence, we need to know the n-2 th element instead of n-1 th . This structure creates an abstract sequence representation, and its acquisition will be based on statistical regularities , which are also fundamental in complex cognitive skills such as the human language (Christiansen & Chater, 2015). The second novel contribution of our study is that we also test the implicit sequence knowledge of our participants after the (instructed) training phase. Our learning task was completed in two different phases. In the first phase, we instructed the participants to focus either on accuracy or speed while performing the task (Different Instruction Phase, Accuracy vs. Speed Group). After the training phase, we tested both groups of participants with the same instruction (i.e., focusing both on accuracy and speed, Similar Instruction Phase). By doing so, we aimed to differentiate between the effects of instructions on the training performance and the acquired knowledge. Our questions were 1) whether the speed/accuracy instruction affects the learning of probabilistic statistical regularities; and if instructions do modify the learning process 2) do they affect the training performance (Different Instruction Phase) and the retrieval of knowledge (Similar Instruction Phase) equally?

Participants
Sixty-six healthy young adults took part in the study. Five of them were excluded from the experiment because they conceivably misunderstood the instructions. Their performance was more than two standard deviations from the mean of their group in more than 50% of the epochs (units of analysis), which was not observable during the practice session. Therefore, 61 participants remained in the final sample (40 females). They were between 19 and 27 years of age (M age = 21.18 years, SD age = 2.13 years). All of them were undergraduate students from Budapest, Hungary (M years of education = 14.14 years, SD years of education = 1.64 years). Participants had normal or corrected-to-normal vision, none of them reported a history of any neurological and/or psychiatric disorders, and none of them was taking any psychoactive medication at the time of the experiment. Handedness was measured by the Edinburgh Handedness Inventory (Oldfield, 1971). The Laterality Quotient (LQ) of the sample varied between −84.62 and 100 (−100 means complete left-handedness, 100 means complete right-handedness; M LQ = 62.25, SD LQ = 53.73). They performed in the normal range on the Counting Span Task (M Counting Span = 3.66, SD Counting Span = 0.81) All of the participants gave written informed consent before enrollment and received course credit for participating. They were randomly assigned to the Accuracy Group (n = 31) or Speed Group (n = 30). No group differences were observed in terms of age, years of education, handedness, and neuropsychological performance (see Table   1). The study was approved by the Research Ethics Committee of the Eötvös Loránd 6 University, Budapest, Hungary, and it was conducted in accordance with the Declaration of Helsinki.

Alternating Serial Reaction Time task
In this study, we used the implicit version of the Alternating Serial Reaction Time (ASRT) task (J. H. Howard & Howard, 1997;). In the ASRT task, four empty circles (300 pixels each) were presented horizontally in front of a white background in the middle of a computer screen. A target stimulus (a drawing of a dog's head, 300 pixels) was presented sequentially in one of the four empty circles ( Figure 1A). A keyboard with four heightened keys (Z, C, B, and M on a QWERTY keyboard) was used as a response device, each of the four keys corresponding to the circles in a horizontal arrangement. Participants were asked to respond with their middle and index fingers of both hands by pressing the button corresponding to the target position. At the beginning of each block of the ASRT task, the four empty circles appeared horizontally on the screen for 200 ms, and then, the first target stimulus occurred, and it remained on the screen until the first correct response. The next stimulus appeared after a 120 ms response-to-stimulus interval.
The serial order of the four possible positions (coded as 1, 2, 3, and 4) in which target stimuli could appear was determined by an eight-element probabilistic sequence. In this sequence, every second element appeared in the same order, while the other elements' positions were randomly chosen out of the four possible locations (e.g., 2r4r3r1r; where r indicates a random position). Therefore, some combinations of three consecutive trials (triplets) occur with a greater probability than others. For example, 2_4, 4_3, 3_1, and 1_2 (where ''_" indicates any possible middle element of the triplet) would often occur because the third element (bold numbers) could be derived from the sequence (or occasionally could be a random element as well). In contrast, 1_3 or 4_2 would occur with less probability because the third element could only be random ( Figure 1B). Therefore, the third element of a high-probability triplet is more predictable from the first event when compared to a lowprobability triplet. There were 64 possible triplets in the task (four stimuli combined for three consecutive trials). Sixteen of them were high-probability triplets, each of them occurring in approximately 4% of the trials, about five times more often than the low-probability triplets.
Overall, high-probability triplets occur with approximately 62.5% probability during the task, while low-probability triplets only occur with a probability of 37.5% ( Figure 1C). As participants practice the ASRT task, their responses become faster and more accurate to the high-probability triplets compared to the low-probability triplets, revealing statistical learning throughout the task (J. H. Howard & Howard, 1997;Kóbor et al., 2017;Song, Howard, & Howard, 2007;Unoka et al., 2017). Each block of the ASRT task contained 85 stimuli (5 random elements at the beginning of the block, then the 8-element alternating sequence repeated 10 times). an eight-item long alternating sequence structure. (B) High-and low-probability triplets. Due to the alternating sequence structure, some runs of consecutive visual stimuli (called triplets) occurred with a higher probability than others. Every trial was defined as the third trial of a high-or a low-probability triplet, based on the two preceding trials. High-probability triplets can be formed by two patterns and one random element, but also by two random and one pattern element. (C) The proportion of high-and low-probability triplets. High-probability triplets occurred in 62.5% of all trials (of which 50% came from pattern trials, i.e., from P-r-P structure, and 12.5% came from random trials, i.e., from r-P-r structure, by chance). Low-probability triplets occurred in the remaining 37.5% of all trials (of which each individual low-probability triplet occurred with a 12.5% probability by chance, originating only from r-P-r structure). (D) The design of the study. In the Different Instruction Phase, different instruction was told to the Accuracy and the Speed Group. After four epochs (each containing five blocks) of the ASRT task, and a 10-minute long rest period, the instruction changed. In the fifth epoch (containing five blocks of stimuli), the same instruction was given to all of the participants (Similar Instruction Phase).

Inclusion-Exclusion Task
We also administered the Inclusion-Exclusion Task (Destrebecqz & Cleeremans, 2001;Destrebecqz et al., 2005;Fu, Dienes, & Fu, 2010;Jiménez, Vaquero, & Lupiáñez, 2006), which is based on the "Process Dissociation Procedure" (Jacoby, 1991). In the first part of the task, we asked participants in what order the stimuli (both pattern and random elements) appeared during the task, and they were asked to type the sequence using the same four response buttons the participants used during the ASRT task (inclusion instruction). After that, they had to generate new sequences that are different from the learned one (exclusion condition). Both parts consisted of four runs, and each run finished after 24 button presses, which is equal to three rounds of the eight-element alternating sequence (Horvath, Torok, Pesthy, Nemeth, & Janacsek, 2018;Kiss, Nemeth, & Janacsek, 2019;Kóbor et al., 2017). The successful performance in the inclusion condition can be achieved by solely implicit knowledge (however, explicit knowledge can also boost performance, but it is not necessary to the successful completion of the task). On the contrary, successful performance in the exclusion condition (i.e., generating a new sequence that is different from the learned one) can only occur if the participant has conscious knowledge about the learned statistical regularities.
Generation of the learned statistical regularities above chance level even in the exclusion task indicates that the participant relies on their implicit knowledge, as it cannot be controlled consciously. To test whether the participants gained consciously accessible triplet knowledge, first, we calculated the percentage of the generated high-probability triplets in the inclusion and exclusion condition separately. Then we tested whether the occurrence of highprobability triplets differs from the probability of generating them by chance. We also compared the percentages of the high-probability triplets across conditions (inclusion and exclusion task) and groups (Accuracy Group and Speed Group) (for more details about the Inclusion-Exclusion task, see: Horvath et al., 2018;Kiss et al., 2019;Kobor et al., 2017).

Questionnaire
We used a questionnaire to scrutinize whether the participants prefer accuracy or speed in general and whether they are rather accurate or fast in their everyday life. The questionnaire consisted of the following questions: "In an everyday situation, what do you attend more: speed or accuracy (in a scale from 1 to 10, where 1 means that only the accuracy is important and 10 means that only the speed is important)?", "In an everyday situation, how important is for you to be accurate/fast in a scale from 1 to 10?", "According to your friends and family, how fast/accurate are you when you need to solve a problem (in a scale from 1 to 10)?".

Design
First, the participants completed three practice blocks of 85 random trials each to familiarize themselves with the task. After that, the participants completed two sessions of the ASRT task. In the first, training session (referred to as Different Instruction Phase), we gave different instructions to the participants of the Accuracy and Speed Group. For the Accuracy Group, the instruction was to try to be as accurate as possible during the task. On the contrary, the instruction for the Speed Group was to be as quick as possible. By this, we could investigate the performance on the task in light of the different instructions. Twenty blocks were presented to the participants in the Different Instruction Phase (for analysis, we organized the blocks into four epochs by merging five consecutive blocks). Participants could rest a bit after each block. A 10 min rest period was inserted before the second ASRT session. During this period, participants were not involved in any demanding cognitive activity. The second session of ASRT (referred to as Similar Instruction Phase) contained five blocks (one epoch).
This time, both the Accuracy and Speed Group were instructed to respond to the target stimulus as quickly and as accurately as possible ( Figure 1D). After the ASRT task, the Inclusion-Exclusion task was administered.

Statistical analysis
We defined each trial as the third element of a high or low-probability triplet. Trills (e.g., 1-2-1) and repetitions (e.g., 1-1-1) were eliminated from the analysis because participants tend to show pre-existing response tendencies to these type of triplets (D. V. Howard et al., 2004; Janacsek, Borbély-Ipkovich, Nemeth, & Gonda, 2018;Takács et al., 2018;Unoka et al., 2017). The first five button presses were random; thus, only the eighth button press could be evaluated as the last element of a valid triplet. Therefore, the first seven trials were excluded from the analysis. Blocks were collapsed into four epochs in the Different Instruction Phase (Epoch 1-4), and one epoch in the Similar Instruction Phase (Epoch 5) to facilitate data processing and to reduce intra-individual variability. We calculated the median reaction times (RTs) separately for high-and low-probability triplets for each participant and each epoch.
Only correct responses were considered for the RT analysis.
We used mixed-design ANOVAs to compare the learning performance between the two groups in the Different and Similar Instruction Phase. In the Different Instruction Phase, we included the factors of Epoch (Epoch 1 to 4), Triplet (high-vs. low-probability triplets), Exclusion) × 2 (Group: Accuracy Group vs. Speed Group) mixed-design ANOVA.
Additionally, we correlated the average RTs and accuracy scores with the rates of the different items of the questionnaire to check whether the subjective preferences of the participant are related to the ability to follow the instructions.

Results
Did the two groups perform equally before learning?
To ensure that there are no pre-existing differences between groups in terms of speed or accuracy, we compared the median RTs (only for correct responses) and the accuracy of the two groups in the practice session. We did not find differences between groups either in RTs, t(59) = -0.48, p = .64, or in accuracies, t(59) = -1.08, p = .28. Therefore, we assumed that there are no pre-existing differences between groups regarding their speed or accuracy. 13

Did the instruction affect general RTs and accuracies?
We ran a mixed-design ANOVA with Epoch (1-4) as within-subject factor and Group

Did the learning process differ between groups as a result of the different instructions?
Next, we investigated whether the sequence learning process differed between groups during the Different Instruction Phase. RTs were analyzed with a mixed-design ANOVA with Triplet

Did the acquired knowledge on the ASRT task differ between groups when testing with the same instructions?
First, we calculated the median RTs separately for the high-and low-probability triplets at the Similar Instruction Phase. We analyzed RTs of Epoch 5 with a mixed-design ANOVA with Triplet (high-probability triplets vs. low-probability triplets) as within-subject factor and with Group (Accuracy Group vs. Speed Group) as a between-subject factor. A significant main

Did the participants develop conscious knowledge about the statistical regularities, and was it different between groups?
The Inclusion/Exclusion task was administered to reveal whether the acquired statistical knowledge remained implicit or became explicitly accessible for the participant. Separately for the two groups, we compared the percentage of the generated high-probability triplets to the chance level (25%). In the Accuracy Group, two participants were excluded from this analysis as they did not follow the instructions. Participants in the Accuracy Group generated 32.33% high-probability triplets in the Inclusion condition, which is significantly higher than chance level, t(28) = 4.82, p < .001. Participants of the Accuracy Group generated highprobability triplets significantly above chance (29.81%) in the Exclusion condition as well t(28) = 4.04, p = .001, indicating that they could not consciously control the emergence of this knowledge. In the Speed Group, two participants were excluded as they did not follow the instructions. Participants of the Speed Group generated 29.34% high-probability triplets in the Inclusion condition, which is significantly higher than chance level, t(27) = 3.58, p = .001.
They also generated more high-probability triplets than expected by chance in the Exclusion condition, 29.25%; t(27) = 2.07, p = .048.
Furthermore, we compared the differences between groups and between tasks with a 2 (Condition: Inclusion vs. Exclusion) × 2 (Group: Accuracy Group vs. Speed Group) ANOVA.
The main effect of Condition was not significant, F(1, 55) = 1.66, p = .20, η p 2 = .03, thus, the triplet knowledge of the participants remained implicit. The Group main effect did not reach significance, F(1, 55) = 0.53, p = .47, η p 2 = .01, indicating that the two groups performed equally on the two tasks. The interaction of the Condition and the Group factors was not significant, F(1, 55) = 0.26, p = .61, η p 2 = .01, meaning that the lack of difference between groups was not influenced by the type of the task.

Did the preferences of the participants affect their performance on the task?
We used a questionnaire to check whether the subjective preferences on being fast or accurate in the real-life were related to the ability to follow instructions (see Methods for the questions). We correlated the questionnaire scores with the average RTs and accuracy of the participants separately for the two groups. We did not find any significant correlations between the average scores and subjective ratings about the preferences either in the Accuracy Group or in the Speed Group (all p > .01). It indicates that the preference of accuracy or speed, and whether the participants are rather fast or accurate in real life did not play a role in the ability to follow the instructions.

Discussion
Here, we aimed to unveil whether speed/accuracy instructions can influence an essential compound of skill learning, namely the acquisition of probabilistic statistical regularities learning. To this end, we instructed two groups of participants to be either fast or accurate during the training in our implicit probabilistic sequence learning task. In the testing phase, we assessed the accumulated knowledge of probabilistic regularities, and this time, all of the participants were instructed to be equally fast and accurate. As predicted, the instructions greatly affected the general speed and accuracy of the participants: the speed instructions resulted in faster reaction times and a higher number of errors, while the accuracy instructions caused slower overall reaction times and an almost errorless performance. Despite these differences during training, the sequence learning index based on RTs was similar in both groups. Thus, the instructions did not affect the acquisition of implicit probabilistic regularities during the training. Moreover, no difference between the groups was found in the 20 testing phase. This lack of difference suggests that instructions did not affect either the performance during training or the acquired statistical knowledge. Similar results were obtained when we controlled for the differences in average speed between groups. Moreover, Bayesian statistical methods also supported the lack of difference between groups in terms of the acquired knowledge.
Our main result is that we detected a similar level of acquired knowledge irrespective of the strategy used during the training. This finding has several implications. From a narrower, learning perspective, it suggests that our ability to extract the relevant pieces of statistical information from the environment is so robust that instructions cannot influence it.
Another compelling result of our study is that participants in the accuracy condition did acquire stable statistical knowledge despite the minimization of motor (response) errors 21 during the training. This statistical knowledge was equal to the knowledge acquired with speed instructions, which was characterized by a relatively high amount of errors during the training. This result is especially interesting in the light of the theory claiming that the brain is a Bayesian inference machine (Friston, 2010) because our results contradict the findings that committing errors facilitates learning (Bubic, Von Cramon, & Schubotz, 2010). Our brain learns associations between events through the continuous adjustments of the estimated probability distribution, i.e., the prior. After a prediction error, the prior should be updated in accordance with the new information about the probabilistic structure (Friston, 2010). Based on these theories, we would expect a low number of errors to impair the learning process.
This was not the case in our study, which raises the possibility that the motor aspect of prediction errors is not crucial in all circumstances for updating the priors during probabilistic sequence learning. However, it is also possible that a similar amount of errors might be detected with other methods, for example, by investigating eye-movements (Le Pelley, Beesley, & Griffiths, 2011;Wills, Lavric, Croft, & Hodgson, 2007). The exploration of the role of errors in implicit sequence learning deserves future investigation, using eye-tracking during sequence learning and electrophysiological methods to measure error-related brain activity.
At the initial training phase, a similar level of statistical learning was found under speed and accuracy instructions. The fact that no group differences were found is in contrast with the results of Hoyndorf and Haider (2009), as they detected impaired implicit performance with accuracy strategy. In their study, participants performed a regular and a random task set during a Number Reduction Task. They found that only the participants focusing on speed had increased speed for the regular task set. The authors claimed that the increased monitoring due to the accuracy instruction might have impeded the performance, similarly as in skill acquisition studies (Beilock et al., 2008(Beilock et al., , 2004. However, in the same study, Hoyndorf and Haider (2009) found a preference for the regular task set also in the accuracy group, which they interpreted as that the focus on accuracy affects only the expression of implicitly acquired knowledge rather than learning processes per se. This is in accordance with our results, as we found a similar level of sequence knowledge when we equally emphasized the importance of speed and accuracy after the initial learning. The difference regarding the training phase might be explained by the fact that we studied more complex, probabilistic sequence representations, which might be more resistant to speed and accuracy strategies than deterministic patterns. Similarly, Barnhoorn et al (2019), who have also found speed instruction to benefit the development of sequence representations, used simple repeating sequences. Moreover, this study investigated explicit sequence learning processes, while our participants were unaware of their accumulated sequence knowledge. A possible explanation for the difference between the effect of implicit and explicit learning conditions could be that the increased speed covers up the explicitness of the task. As a consequence, the task becomes more implicit, the top-down control reduces, and the learning becomes better. In our study, the learning was entirely implicit, therefore, the speeding up could not improve the level of implicitness. Thus, the learning was similar under speed and accuracy instructions. Future investigation is needed to reveal to what extent is the implicit or the probabilistic nature of the task related to lack of speed benefit during training. should also be tested, as general speed-up and changes in accuracy can be seen over the course of several cognitive tasks requiring fast decision-making. Based on our results, we recommend taking into consideration the possible differences between the measured competence and performance when designing learning studies.
We manipulated the general speed and accuracy of the participants by giving explicit instructions to focus either on speed or accuracy, as previous non-learning cognitive tasks also did (e.g., Aasen & Brunner, 2016;Christensen et al., 2001;Osman et al., 2000;Ullsperger, Bylsma, & Botvinick, 2004). However, it might be questionable if our results truly reflect the effect of instructions on learning. One can argue that the instructions given in our study were not strong enough to manipulate the learning strategy and therefore, the learning processes because previous studies on the topics used more pronounced instructions and signals to modify the strategy of the participants (Barnhoorn et al., 2019;Hoyndorf & Haider, 2009).
This seems unlikely as the general speed and accuracy were affected by the instructions.
Group differences also emerged in general skill learning as (1) participants who focused on their speed showed increasingly faster responses, and (2) participants who focused on their accuracy sustained a high level of accuracy during the learning phase compared to the other group. In contrast to these findings, the acquisition of statistical regularities was not affected by the instructions. To sum up, we found evidence that speed and accuracy affect general skill learning and sequence-specific learning (statistical learning) differently.
It can also be claimed that verbal instructions given at the beginning of the task might not be sufficient to regulate subjects' average speed and accuracy, because as time goes on, participants tend to wane in favor of their response tendencies (Heitz, 2014). In other words, they will behave according to their preferences for being accurate or fast on a task. In our case, this change in behavior is unlikely. First, we found no differences in the average RTs and accuracy scores between groups when the participants practiced the task on random sequences (before we gave distinct instructions to the groups), and second, participants did not become less accurate or slower throughout the task. Therefore, the effects observed should be the results of the instructions. Additionally, we measured the participants' individual preferences on response tendencies with a questionnaire (whether they prefer to be accurate or fast). No correlations were observed between these individual preferences and the average speed and accuracy during the task in either group. These aspects indicate that our results indeed reflect the effect of instructions, and participants did not follow their individually preferred response tendencies during the task.

Conclusions
Our study investigated the effects of speed and accuracy instructions on an essential compound of skill learning, namely the acquisition of probabilistic regularities. Our main finding is that our ability to pick-up statistical regularities in a noisy, uncertain environment is 25 so robust that instructions do not influence it. It indicates that implicit probabilistic sequence learning is independent of the manipulation of speed/accuracy trade-off. Another finding of our study is that the learning is intact with almost 100% accuracy level. It suggests that statistical learning is at least partly independent of accuracy level, and statistical knowledge about the environmental regularities can be acquired even if no response (motor) errors occur.
Our results also raise the possibility that competence and performance can differ in some instances. Accuracy instructions can mask the accumulating knowledge during learning when measured by accuracy, although statistical knowledge does emerge in these cases as well.
Future studies investigating whether this robustness is related to the implicit feature of the task or whether different types of learning are affected equally, seem warranted.