Fig. 2: Reconstructing a map-like representation from the behavioural data.

The estimated pairwise similarity scores of action combinations (A) and coloured balls (C) from the Rating Tasks were correlated with the distances between their positions in the action-outcome space (Task 1: M = 0.451, SD = 0.288, Spearman’s rho; t(45) = 10.63, p < 0.001, Cohen’s d = 1.56, 95% CI= [0.888, 2.246]; Task 2: M = 0.389, SD = 0.27, Spearman’s rho; t(45) = 9.79, p < 0.001, Cohen’s d = 1.44, 95% CI= [0.776, 2.11]; two-sided t-test). Dots indicate data from individual participants. For both tasks, this correlation was higher on the second day compared to the first day (Task 1: M = 0.604, SD = 0.282, Spearman’s rho; t(45) = 14.52, p < 0.001, Cohen’s d = 2.14, 95% CI= [1.396, 2.886]; two-sided t-test; Across-day effect: t(45)= −4.28, p < 0.001, Cohen’s d = −0.535, 95% CI= [−0.801, −0.269]; two-sided paired t-test; Task 2: M = 0.571, SD = 0.261, Spearman’s rho; t(45) = 14.84, p < 0.001, Cohen’s d = 2.18, 95% CI= [1.437, 2.939]; two-sided t-test; Across-day effect: t(45)= −4.82, p < 0.001, Cohen’s d = −0.686, 95% CI= [−1.001, −0.372]; two-sided paired t-test). B, D To visually evaluate how participants’ responses reflected the structure of the action-outcome space, we reconstructed their implied mental space from the pairwise similarity scores by applying multidimensional scaling (MDS) and obtained coordinate positions (indicated by snowflakes) that represent the structure of these similarity estimates. The resulting coordinates were mapped using Procrustes analysis to match the original positions (indicated by circles) of action combinations or coloured balls in the action-outcome space (see Bellmund et al.¹⁰⁷ and see Methods). The corresponding left and middle panels demonstrate the data of an example participant from day 2. To evaluate the significance of mapping, we fitted MDS coordinates using the same approach to the sets of coordinates with shuffled stimulus – position assignments (curved solid lines). The corresponding right panels show the data for all participants. The vertical solid lines represent the mean Procrustes distances between original and reconstructed positions for each training day. The vertical dashed lines indicate mean critical Procrustes distances obtained from the shuffled distributions for each training day. The shaded areas indicate the standard error of the mean, calculated across participants. The results revealed that the mapping was indeed better than would be expected by chance for action combinations (Day 1: data distance: M = 0.368, SD = 0.273; critical distance: M = 0.526, SD = 0.009; Day 2: data distance: M = 0.257, SD = 0.261; critical distance: M = 0.528, SD = 0.011) as well as coloured balls (Day 1: data distance M = 0.371, SD = 0.219; critical distance: M = 0.382, SD = 0.028; Day 2: data distance: M = 0.247, SD = 0.205; critical distance: M = 0.372, SD = 0.04), pointing towards organization of multiple action-outcome associations in an abstract two-dimensional map. A, C Dots represent data from n  =  46 participants. boxplots show median and upper/lower quartile with whiskers extending to the most extreme data point within 1.5 interquartile ranges above/below the quartiles; black circles with error bars correspond to mean ± SEM; distributions depict probability density functions of data points. Source data are provided as a Source Data file. ****p < 10^{^−4}; Bonferroni corrected for tests on both days.