Comparing Offertory Melodies of Five Medieval Christian Chant Traditions

In this study, we compare the melodies of five medieval chant traditions: Gregorian, Old Roman, Milanese, Beneventan, and Mozarabic. We present a newly created dataset containing several hundreds of offertory melodies, which are the longest and most complex within the total body of chant melodies. For each tradition, we train ngram language models on a representation of the chants as sequence of chromatic intervals. By computing perplexities of the melodies, we get an indication of the relations between the traditions, revealing the melodies of the Gregorian tradition as most diverse. Next, we perform a classification experiment using global features of the melodies. The choice of features is informed by expert knowledge. We use properties of the intervallic content of the melodies, and properties of the melismas, revealing that significant differences exist between the traditions. For example, the Gregorian melodies contain less step-wise intervals compared to the other repertoires. Finally, we train a classifier on the perplexities as computed with the n-gram models, resulting in a very reliable classifier.