We remember what we like?: Aesthetic value and memorability for photos and artworks - a combined behavioral and computational study
23.4014, Saturday, 16-May, 8:30 am - 12:30 pm, Pavilion
Christian Wallraven1, Joern Freese1; 1Cognitive Systems, Brain and Cognitive Engineering, Korea University
Several studies testing long-term pictorial memory have claimed nearly unlimited memory capacity even for large amounts of stimuli. Additionally, theories in empirical aesthetics have suggested that the aesthetic appreciation of a stimulus and its memorability may be linked – “we remember what we like”. In the present study we investigated these questions for two types of stimuli: photos and artworks. Specifically, we tested whether the memorability of a photo or artwork may be connected to its aesthetic value. A total of 52 participants took part in a two-phase experiment. In the first phase, 750 images (photos of everyday scenes, Korean artworks, and Western artworks) had to be memorized and rated for aesthetic value. In the second phase one day later, these images had to be recognized in an old-new paradigm. Overall, participants’ long-term memory capacity proved to be significantly more limited than suggested before (average d-prime=0.987). Nonetheless, memorability scores themselves were reliable across participants. Surprisingly, artworks – although appreciated more aesthetically – were less memorable (d-prime=0.854) than everyday scenes (d-prime=1.201). Furthermore, aesthetic ratings correlated only marginally with memorability (r=0.174), suggesting that these two variables need to be examined separately in future research and modeling. For our computational study, we then tested a large set (n=1025) of pictorial, low-level image features in terms of correlations with the behavioral results. The features were able to predict the correct stimulus class with 85% accuracy (chance=33%). Interestingly, we found small, yet significant correlations for some low-level measures for aesthetic ratings (maximum r=0.312), but not for memorability scores (maximum r=0.15). Overall, however, computational features only provided a small amount of explanatory power. Taken together, these results show that aesthetic appreciation and memorability are two independent concepts (for both art and everyday scenes). In addition, memory representations and aesthetic experience seem driven largely by high-level interpretation processes.