Moral reframing is the process of framing a statement in a way that is consistent with an individual's moral values to garner support from that individual. Prior studies have shown morally-reframed messages help in persuasion on a range of topics, from environmental protection to reducing political polarization. We developed a moral classifier and a generator to morally reframe texts, which is the first NLP approach to do so. Specifically, we developed a moral classifier that can predict the underlying morals in a text using a BART-based architecture and a moral reframer that can perform moral reframing of an input text based on a target set of morals using a variational autoencoder architecture. Our best moral reframer model achieves F1 scores of 0.998 on our test set and a BERTSCORE (using BART) of 0.972, thus being able to perform moral reframing while preserving content.
