BootTOD proposes a self-bootstrapping framework for task-oriented dialogue representations, outperforming contrastive methods and enhancing diversity in responses.
