Procedural learning, crucial for skill acquisition in daily life, is influenced by instructions like prioritizing speed or accuracy. Prior research yielded contradictory results, possibly because procedural learning encompasses diverse cognitive functions. This study aims to clarify this by examining how speed and accuracy instructions affect probability-based and serial-order-based regularities in procedural learning. Probability-based learning involves acquiring short-range associations between sequence elements based on their frequency, while serial-order-based learning involves learning the sequence of events. These learning aspects are conceptually and methodologically distinct, characterized by different neural mechanisms and developmental trajectories. Probability-based learning tends to be incidental and rapid, leading to robust representations, whereas serial-order-based learning can be either incidental or intentional and takes longer to develop. We hypothesize that the incidental, rapid learning of probability-based regularities is more susceptible to instruction manipulation, predicting that speed instructions will enhance probability-based learning based on previous findings showing better procedural learning under speed instructions. Our previous work using a non-cued version of the Alternating Serial Reaction Time (ASRT) task showed no effect of speed or accuracy instructions on procedural knowledge. Using a cued version of the ASRT task in this study allows for the dissociation of probability-based and serial-order-based learning within a shorter timeframe, enabling us to investigate the effects of instructions more effectively.

Studies on skill acquisition in sports suggest that novices benefit from accuracy instructions, while experts benefit from speed constraints. Hoyndorf and Haider (2009) found that accuracy instructions impaired implicit skill acquisition compared to speed instructions, although learning occurred under accuracy instructions. Barnhoorn et al. (2019) showed that speed instructions enhance the representation of explicit sequences, while accuracy instructions lead to faster response selection through better stimulus-response associations. However, our previous work indicated that procedural knowledge acquisition and retention were not affected by speed or accuracy instructions. This inconsistency highlights the multifaceted nature of procedural learning, with different learning mechanisms potentially contributing differently to the overall outcome. This study aims to resolve this by focusing on the distinct aspects of probability-based and serial-order-based learning.

Fifty-six healthy young adults participated; eight were excluded due to inconsistent performance, leaving 48 (43 female) participants. Participants were randomly assigned to either a Speed Group or an Accuracy Group. The cued version of the ASRT task was used, where participants responded to visual cues (dog head or penguin) appearing in one of four locations using corresponding keyboard keys. The task included a predetermined sequence (pattern trials) interspersed with random trials. Pattern trials were visually distinct (dog heads), while random trials were visually distinct (penguins). This alternating structure created high- and low-probability triplets of trials. Probability-based learning was measured by comparing reaction times (RTs) for random high-probability and random low-probability trials. Serial-order-based learning was measured by comparing RTs for pattern high-probability and random high-probability trials. Participants first completed three practice blocks, followed by 20 blocks with instructions to focus on either speed or accuracy (Different Instruction Phase). After a 10-minute break, participants completed five more blocks with instructions to focus on both speed and accuracy (Same Instruction Phase). After each block, participants performed a sequence report task to assess their explicit knowledge of the pattern sequence. Median RTs were calculated for each trial type and standardized by the median RT for that epoch to account for the speed differences between groups. Mixed-design ANOVAs were used to analyze RT data for both phases, and Mann-Whitney U tests for the Same Instruction Phase. Bayesian ANOVAs and Mann-Whitney U tests were also conducted.

Before the learning phase, no differences were found in RT or accuracy between the two groups. In the Different Instruction Phase, the Speed Group was faster and less accurate than the Accuracy Group, as expected. The Speed Group showed significantly better probability-based learning scores than the Accuracy Group in the Different Instruction Phase. However, in the Same Instruction Phase, both groups showed comparable probability-based learning. There were no significant differences in serial-order-based learning between groups in either phase. In the sequence report task (explicit knowledge), the Accuracy Group showed a trend-level advantage in the first epoch of the Different Instruction Phase, but both groups showed comparable performance in later epochs. In the Same Instruction Phase, both groups again demonstrated comparable performance. The study shows that the speed instructions mainly affected the expression of probability-based knowledge during the initial learning phase, but not in the subsequent phase. There were no significant differences in learning or acquired knowledge in serial-order-based learning, which seems resistant to the instructions manipulation of the speed/accuracy trade-off.

This study shows that speed and accuracy instructions differentially affect probability-based and serial-order-based learning. Speed instructions enhance the expression of probability-based knowledge but don't affect acquired knowledge. Serial-order-based learning appears resistant to instructions. These findings are consistent with Hoyndorf & Haider (2009) who also suggested that the effects of instructions are mostly on performance rather than learning itself. The difference between the current and previous findings (using the non-cued version) might be attributed to the cueing in the current task. The results highlight the distinction between momentary performance and acquired knowledge (competence), suggesting that using single-session evaluations may be insufficient to assess long-term probability-based learning. The almost error-free learning under accuracy instructions suggests that response errors are not essential for procedural learning, challenging error-driven learning theories. However, more errors in the speed condition might be initially advantageous but do not provide long-term advantage. The observed trend-level advantage of the Accuracy Group in explicit knowledge at the beginning of the task suggests that accuracy instructions promote explicit learning.

Speed and accuracy instructions differentially affect probability-based and serial-order-based regularities in procedural learning. Speed instructions benefit the expression of probability-based knowledge, but not the acquired knowledge. Serial-order-based learning remains unaffected. This study highlights the importance of distinguishing between momentary performance and acquired knowledge and that procedural learning does not exclusively depend on response errors. Future research could explore the interaction between instruction type, learning type (implicit/explicit), and gender differences.

The study lacked a control group, which could have helped mitigate potential biases due to the cued nature of the task. The sample was predominantly female, potentially limiting the generalizability of findings regarding gender differences. The study did not use an imaging technique, hindering exploration of the relationship between neural activity and learning under different instructions.