BERTian Poetics: Constrained Composition with Masked LMs

Abstract

Masked language models have recently been interpreted as energy-based sequence models that can be generated from using a Metropolis--Hastings sampler. This short paper demonstrates how this can be instrumentalized for constrained composition and explores the poetics implied by such a usage. Our focus on constraints makes it especially apt to understand the generated text through the poetics of the OuLiPo movement.

Accepted as a poster at the 2021 NeurIPS Workshop on Machine Learning for Creativity and Design