Encoding Notre Dame polyphony for corpus analysis

Authors: Stutter, Joshua

Date: Wednesday, 6 September 2023, 11:15am to 12:45pm

Location: Main Campus, L 2.202 <campus:measure>


The encoding of 13th-century vocal music, known as Notre Dame (ND) polyphony, presents novel challenges for music encoding. Beyond the peculiarities of its notation, including unwritten notes, diverging variation in sources, and a subtle interplay of musical reuse between the two types of notations for organa and motet, this paper argues that the main challenges facing a satisfactory encoding of ND polyphony are: a) the representation of modal rhythm, and b) a lack of academic consensus concerning certain basic aspects of the notation.

Unlike unmeasured monophony such as plainchant or fully rhythmically-specified polyphony such as mensural notation or CWMN, an encoding of ND polyphony cannot be structured by voice leading or duration. The highly idiosyncratic and contextual ligature patterns of ND notation can seldom be interpreted using the system of rhythmic modes: it requires the rules to be bent or broken or, in most cases, left to the discretion of a performer. As a result, the alignment of voices becomes difficult to encode and there are numerous “correct” answers. The only clues are in the differing graphical layouts of the scribes.

Moreover, despite a century of musicological research into the ND repertory, the academic community has yet to reach a common consensus on some of its most fundamental issues. This is due, in part, to the paucity of surviving sources and contemporary writing, the influence of oral traditions on the transmission of sources, and the repertory’s exploitation by the shifting narratives of music history as the so-called “crucible” of Western art music. Typical approaches to encoding ND polyphony must therefore shoehorn a transitional style into either a “neumatic” or “mensural” context when it in fact belongs to neither.

To address these challenges, this paper proposes a new approach to encoding ND polyphony as part of the “Clausula Archive of the Notre Dame Repertory” (CANDR). The approach is based on a two-layer directed graph that avoids neumatic or mensural structures. By expanding the graph into a complete graph using a simple distance metric, this paper demonstrates the building of node embeddings and continuous-bag-of-words embeddings to vectorize sections of music, as well as the training of a support vector machine (SVM) to extract passages of interest. This approach offers a viable solution for addressing the challenges presented by this unique repertory.

Contribution Type