Abstract
There is debate about how implicit and explicit processes interact in sensorimotor adaptation, implicating how error signals drive learning. Target error information is thought to primarily influence explicit processes, therefore manipulations to the veracity of this information should impact adaptation but not implicit recalibration (i.e. after-effects). Thirty participants across three groups initially adapted to rotated cursor feedback. Then we manipulated numeric target error through knowledge of results (KR) feedback, where groups practised with correct or incorrect (+/-15°) numeric KR. Participants adapted to erroneous KR, but only the KR + 15 group showed augmented implicit recalibration, evidenced by larger after-effects than before KR exposure. In the presence of sensory prediction errors, target errors modulated after-effects, suggesting an interaction between implicit and explicit processes.
Author Contributions
BCL and NJH were involved in the conceptualization of the project, study hypotheses, and experimental methods. BCL collected data and completed data analysis. BCL and NJH both contributed to writing of the manuscript and both authors read and approved the final version.
Acknowledgements
The authors thank Romeo Chua for his insightful comments and feedback during the planning of this experiment and writing of the final manuscript.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 A fourth group (n = 5) of participants was later tested under the same testing conditions but no numeric KR was provided during Adapt 2. In Figures 2, 5 and 6 we have added this group to visually demonstrate the similarity of this group to the Correct KR group. Despite the low sample size, we ran an exploratory statistical analyses to compare these groups in a 2 Group X 2 Block (1 vs. 10) LMER for Adapt 2 CE, where both the group-level contrast (ß = -1.58, 95% CI [-5.50 – 2.35], p = .41) and the Group X Block interaction were not statistically significant (ß = 2.02, 95% CI [-2.10 – 6.15], p = .32). There were also no significant differences between the No KR and Correct KR groups in the patterns of errors for these groups in measures of after-effects after Adapt 2 (ßGroup = .44, 95% CI [-4.25 – 5.12], p = .85; ßGroup X Testing Phase = .94, 95% CI [-2.12 – 3.99], p = .54), and in Retention (ßGroup = -.14, 95% CI [-2.80 – 2.51], p = .91; ßGroup X Block = .25, 95% CI [-2.39 – 2.88], p = .85). For full analysis breakdown see Supplementary Analyses, Tables 8-10).
2 The linear model was chosen for all mixed effects analyses. Model fit was evaluated using Akaike’s Information Criterion (AIC). For all analyses involving CE and VE, inclusion of the 3-way interaction produced the lowest AIC compared to the base model, with the exception of analysis for RT. However, due to predictions and to be consistent across measures, we kept the interaction term in the model for all analyses. All fixed effects models were then compared to the mixed effects’ model that included participant as a random effect. All mixed effects models produced the lowest AIC. For crossed random effects we specified a random intercept for each participant within each testing phase and block. In two cases (analysis of adaptation CE and VE data) the term for the random effect of participant within block was removed, because it produced random effect variance estimates of zero. This was informed by recommendations to fit the most complex model that permits a non-singular fit (Barr et al., Citation2013).