Developing a Musicality Scale for Haiku-Likes

Abstract

Musicality is the potential of a poem to become a song. The ability to select a musical poem from a collection of poems is an essential skill for a musician, one that is assumed to be acquired through the practice of dealing with lyrics for years. Answers to the question of what makes a poem musical calls for research that, to the best of our knowledge, has not been addressed in the literature. The computational model discussed in this paper is for checking a poetic piece for its musicality from the perspective of Hindustani Classical Music (HCM). The length of lyrics selected for an HCM performance has reduced over the centuries. The trend may continue, and we shall see tiny poems in the place of today’s 2 to 4 full lines. The case considered here is a novel type of composition, Haiku-bandish (HB), with the potential of setting classical music to the succinct Japanese classic poetry form known as Haiku. We found that a few of the assumptions about what makes a poem musical are not found to be statistically verifiable. The sequence information about the long and short notes at the beginning and at the end is a better choice of feature over the bag of symbols model that considers the information about alliteration, word-length and occurrences of long and short notes. The testing set performance demonstrates that accuracy and F-score are both 85%. A unique byproduct of this research is a set of 16 HB in Marathi, the language of Maharashtra state of India, to which Raaga music has been set by following the time-conventions of HCM.