Diffusion-LM Improves Controllable Text Generation
This addresses the open problem of controllable text generation for natural language processing, offering a novel method for fine-grained control beyond simple attributes.
The paper tackles the problem of controlling language models for complex, fine-grained text generation without re-training, and introduces Diffusion-LM, a non-autoregressive model based on continuous diffusions, which significantly outperforms prior work on six challenging control tasks.
Controlling the behavior of language models (LMs) without re-training is a major open problem in natural language generation. While recent works have demonstrated successes on controlling simple sentence attributes (e.g., sentiment), there has been little progress on complex, fine-grained controls (e.g., syntactic structure). To address this challenge, we develop a new non-autoregressive language model based on continuous diffusions that we call Diffusion-LM. Building upon the recent successes of diffusion models in continuous domains, Diffusion-LM iteratively denoises a sequence of Gaussian vectors into word vectors, yielding a sequence of intermediate latent variables. The continuous, hierarchical nature of these intermediate variables enables a simple gradient-based algorithm to perform complex, controllable generation tasks. We demonstrate successful control of Diffusion-LM for six challenging fine-grained control tasks, significantly outperforming prior work.