Machine learning of symbolic compositional rules with genetic programming: dissonance treatment in Palestrina

Area of Science:

• Computational Musicology
• Music Theory
• Artificial Intelligence

Background:

• Algorithmic composition requires symbolic rules.
• Extracting these rules from music corpora is challenging.
• Genetic programming (GP) offers a potential solution for rule extraction.

Purpose of the Study:

• To develop and evaluate a method for automatically extracting symbolic compositional rules from music.
• To apply this method to model dissonance treatment in Palestrina's music.
• To investigate the efficacy of GP for learning logic and numeric relations in music.

Main Methods:

• Developed a custom algorithm for dissonance labeling in musical fragments.
• Utilized DBSCAN for automatic clustering of melodic fragments into dissonance categories.
• Employed genetic programming (GP) to learn rules for dissonance treatment based on positive and negative examples.

Main Results:

• Successfully extracted symbolic rules combining logic and numeric relations from music corpora.
• Demonstrated the ability of GP to learn rules for specific dissonance categories.
• The learned rules are interpretable by humans and usable for algorithmic composition.

Conclusions:

• The proposed method effectively extracts symbolic compositional rules from music.
• Genetic programming is a viable technique for learning complex musical patterns, specifically dissonance treatment.
• The extracted rules can enhance algorithmic composition systems and musicological analysis.