Expressive Communication: A Commmon Framework for Evaluating Developments in Generative Models and Steering Interfaces
In the Expressive Communication framework, Alice the composer is tasked with composing a piece of music that expresses the imagery and words of a card. She creates two pieces of music, one for each of the different systems she is comparing, which can differ on their interfaces or their models. After completing her compositions, she provides self-report answers about her experience and the compositions made with both systems:
- Expression: I feel confident that my composition created with System X expresses the imagery/ideas of the card I chose.
- Communication: I feel confident that others will be able to rank the Card I chose high after listening to the composition created with System X.
- Musical Coherence: The composition made with System X feels musically coherent.
- Ownership: Using System X, I felt the composition created was mine.
- Control: I felt I had control creating the composition when using System X.
- Efficacy: Using System X, I could find several solutions to achieve my goals for the composition.
Afterwards, Bob the listener is provided the same card and listens to both music compositions made by Alice. Bob is asked to compare these compositions along two measures:
- Which better evokes the imagery and words of the card?
- Which sounds more musical?
Composer Results
Providing composers with more steerable interfaces to the generation helped them feel more ownership, control, and efficacy in finding multiple music options that achieved their goals
Providing composers with with the more expressive model helps them feel the composition has higher musical coherence, and helped them find more viable options in the generated music which ultimately contributed to a small but significant difference in ownership and control.
