Engineering

Project

Gradus ad Parnassum

An early exploration of retrieval over musical notation, working toward answers grounded in specific measures of a score.

The question

What would it take to ask a score a musical question and get an answer that points to the relevant measures? For example: where does Chopin bring out a left-hand melody against the right, or use a deceptive cadence at the end of a phrase?

The approach

The project explores parsing MusicXML and kern scores, annotating passages with musical structure, and retrieving them in response to a question. The initial corpus target is the Chopin Etudes. My role as a pianist informs the questions and the measure references used to evaluate the answers.

Current stage

This is early work. The intended evaluation compares retrieved passages with a hand-written set of questions and known measure answers. Results have not yet been published. The longer-term goal is notation-native generation grounded in retrieval.