The role of art knowledge training on aesthetic judgements and executive functions

Abstract

The study of how we develop art knowledge can provide valuable insights into the underlying cognitive systems that support expertise and knowledge transfer to new contexts. An important and largely unanswered question is whether art knowledge training impacts subsequent judgements of artworks and executive functions. Across three pre-registered experiments (N > 630 total), which used a training intervention and Bayesian regression modelling, we explore whether art knowledge training impacts subsequent judgements of artworks and executive functions. Experiments 1 and 2 revealed an effect of art training on aesthetic judgements for trained but not untrained artworks. These training effects were generalized to unseen artworks produced by the same artist (Experiment 1) or another artist with a similar style (Experiment 2), but not to different art styles. Experiment 2 also showed that with larger training 'doses' (>16 minutes), the generalization effects are stronger. Experiment 3 showed invariance of the attentional network to art training versus non-art training, suggesting similar sensitivity of executive functions to different types of training. This work shines new light on the cognitive systems that support learning and generalization of learning to new contexts. Likewise, from an applied perspective, it emphasizes that learning and generalization can occur rapidly with a relatively short (approx. 16 minutes) training video.

Keywords: aesthetic judgements; art knowledge; dose–response; executive functions; generalization effects; training.

Figure 1.

Categories of art stimuli used in Experiments 1 and 2, across pre- and post-art training. In Experiment 1, the pre-art training involved the presentation of 20 Realism stimuli and 20 Impressionist stimuli by Sorolla. The post-art training included the art stimuli from the pre-training (20 Realism stimuli and 20 Impressionist stimuli by Sorolla) and also 20 new art stimuli by Sorolla and 20 new Post-Impressionist art stimuli by Gauguin. Stimuli for Experiment 2 were similar to Experiment 1 with one exception. The post-art training also included 20 art stimuli by Merritt-Chase.

Figure 2.

A visual description of the order of the tasks in Experiment 1. Both pre-training and post-training consisted of ratings on aesthetic preference, understanding, affective judgement and artistic skill judgement.

Figure 3.

Example of experimental trial during pre- and post-training.

Figure 4.

Ratings across pre- and post-art training on Realism and Sorolla images for all four DVs. The left panel shows the pre- and post-training ratings for Realism, whereas the right panel shows the pre- and post-training ratings for Sorolla. The four rows illustrate our main DVs—from the top row for aesthetic preference, then understanding, to the bottom rows that show the ratings for affect and artistic skill. The ratings are reported on a 5-point Likert scale (1 = not at all to 5 = extremely). Error bars represent 95% confidence intervals. The black markers (circles and triangles) and interval estimates represent the group mean average, whereas the grey markers represent the individual participants.

Figure 5.

Multivariate parameter estimates for the full model (Model 9) across all four dependent variables: preference, understanding, affect and artistic skill. Note: training = pre versus post; image_type = image category (Realism versus Sorolla); training*image_type = interaction between training (pre versus post) and image_type (Realism versus Sorolla). Point estimate = median; error bars represent 66% quantile intervals (thick black lines) and 95% quantile intervals (thin black lines).

Figure 6.

Ratings across generalization conditions (pre-Sorolla compared with post-Sorolla, post-Sorolla new and post-Gauguin) for all four DVs. The top panels show the generalization ratings for aesthetic preference (left) and understanding (right), whereas the bottom panels show the ratings for affect (left) and artistic skill (right). The ratings are reported on a 5-point Likert scale (1 = not at all to 5 = extremely). Error bars represent 95% confidence intervals. The black markers (circles, triangles, squares, crosses) and interval estimates represent the group mean average, whereas the grey markers represent the individual participants.

Figure 7.

Multivariate parameter estimates for the full model (Model 7) across all four dependent variables: preference, understanding, affect and artistic skill. Note: post_Sorolla = artworks by Sorolla presented during post-training; post_Sorolla_new = previously unseen artworks by Sorolla presented during post-training; post_Gauguin = previously unseen artworks by Gauguin presented during post-training. Point estimate = median; error bars represent 66% quantile intervals (thick black lines) and 95% quantile intervals (thin black lines).

Figure 8.

Ratings across training (pre- versus post-), image type (Realism versus Sorolla) and dose (one, two, three, four) on all four DVs. The columns show all four art training doses, whereas the rows show the ratings for aesthetic preference, understanding, affect and artistic skill. The ratings are reported on a 5-point Likert scale (1 = not at all to 5 = extremely). Error bars represent 95% confidence intervals. The black markers (circles and triangles) and interval estimates represent the group mean average, whereas the grey markers represent the individual participants.

Figure 9.

A visual description of the order of the tasks in Experiment 2. Both pre-training and post-training consisted of ratings on aesthetic preference, understanding, affective judgement and artistic skill judgement.

Figure 10.

Multivariate parameter estimates for the full model (Model 13) across all four dependent variables: preference, understanding, affect and artistic skill. The highlighted panels show the main parameters of interest. Panel (A) = average effect of training; Panels (B–D) = three-way interaction between training, image type and dose. Note: training = pre versus post; image_type = image category (Realism versus Sorolla); dose = one—5:03 minutes; two—10:18 minutes; three—16:04 minutes; four—22:08 minutes. Point estimate = median; error bars represent 66% quantile intervals (thick black lines) and 95% quantile intervals (thin black lines).

Figure 11.

Ratings across dose type and generalization conditions (pre-Sorolla compared with post-Sorolla, post-Sorolla new, post-Gauguin and post-Merritt-Chase) on all four DVs (aesthetic preference, understanding, affect and artistic skill). The columns illustrate the art training doses. The rows show the ratings for aesthetic preference, understanding, affect and artistic skill. The ratings are reported on a 5-point Likert scale (1 = not at all to 5 = extremely). Error bars represent 95% confidence intervals. The black and interval estimates represent the group mean average, whereas the grey markers represent the individual participants.

Figure 12.

Multivariate parameter estimates for the full model (Model 9) across all four dependent variables: preference, understanding, affect and artistic skill. The highlighted panels show the main parameters of interest. Panels (A–D) = average effect of condition. Panels (E) and (F) = two way interaction between post_sorolla condition and dose. Note: post_Sorolla = artworks by Sorolla presented during post-training; post_Sorolla_new = previously unseen artworks by Sorolla presented during post-training; post_Gauguin = previously unseen artworks by Gauguin presented during post-training; post_Merritt_Chase = previously unseen artworks by Merritt-chase presented during post-training; dose = one—5:03 minutes; two—10:18 minutes; three—16:04 minutes; four—22:08 minutes. Point estimate = median; error bars represent 66% quantile intervals (thick black lines) and 95% quantile intervals (thin black lines).

Figure 13.

Example of Attentional Network Test (ANT) experimental conditions. Upper panel (A) shows the three flanker types: neutral, congruent, incongruent; lower panel (B) shows the four cue types: no cue, central cue, double cue, spatial cue.

Figure 14.

A visual description of the order of the tasks in Experiment 3.

Figure 15.

Violin plots on summary data showing ANT reaction time across all conditions. Reaction time is reported in milliseconds (ms). The columns show response times on art versus non-art across pre- and post-training, whereas the rows show the DVs (alerting, executive, orienting). Error bars represent 95% confidence intervals. The black markers (circles and triangles) and interval estimates represent the group mean average, whereas the grey markers (circles and triangles) represent the individual participants.

Figure 16.

Parameter estimates for each predictor within the full model (Model 17). The highlighted panels show the main parameters of interest. Panels (A–C) = three way interaction effects of session, group and ANT. Note: group (art versus non-art); session (pre- versus post-training). Point estimate = median; error bars represent 66% quantile intervals (thick black lines) and 95% quantile intervals (thin black lines).