The Relationship Between Musical Structure and Emotion in Classical Piano Scores : A Case Study on the Theme of La Folia

We explored the relationship between musical structure and emotion on different variations of La Folia – a musical theme of Portuguese origin based on a standard harmonic progression. Our approach aims to extend previous research by investigating more factors and comparing different models for music emotion. In a pilot study, 12 participants rated the emotion associated to the stimuli on a graduate scale from 1 to 10, according to 3 different models for music emotion: the valence/arousal-based emotion model (Russell, 1980), a discrete emotion approach (Izard, 1972), and the Geneva Emotional Music Scale (GEMS) (Zentner et al., 2008). Stimuli were commercial recordings of the first 8 bars of 32 variations of the Theme of La Folia by A. Scarlatti, C. P. E. Bach, S. Rachmaninov and F. Liszt, with different combinations of 9 factors that were judged and averaged by 2 musical experts using a 5-point rating scale. Preliminary results are: (i) there exist significant correlations between structural parameters and descriptors for emotions in all of the models; (ii) correlations between structure and emotion are more remarkable for the valence/arousal-based emotion model and for the GEMS model, and are higher for register, note density, dynamics, accentuation and articulation; (iii) agreement among raters for the DES model is significantly lower than for the other two models. On the base of these results, we planned two new experiments focussing only on the valence/arousal-based emotion model and on the firstorder GEMS model. They are: a second listening test based on a real music design (20 out of the previously selected 32 stimuli, corresponding to pieces featuring more extreme variability in the musical structure, with different combinations of 21 factors), and a new listening test based on a controlled and balanced design (24 variations with selected combinations of 4 factors, arranged and recorded in a deadpan performance by the first author). 24 participants (12 musicians and 12 non-musicians) have been involved in the current study, following the same procedure outlined above. We expect to extend understanding on the relationship between structure and emotion following more accurate, computational analyses of musical features, as well as by defining new predictors for emotion that are more appropriate for specific musical style(s) – like factorial analysis or multidimensional scaling.