Style as Sentiment Versus Style as Formality: The Same or Different?

Abstract

Listen

Translate article

Unsupervised textual style transfer presupposes that style is a coherent and consistent concept and that style transfer approaches will generalise consistently across different domains of style. This paper explores whether this presupposition is appropriate for different types of style. We explore this question by comparing the performance and latent representations of a variety of neural encoder-decoder style-transfer architecture when applied to sentiment transfer and formality transfer. Our findings indicate that the relationship between style and content shifts between these different domains of style: for sentiment, style and content are closely entangled; however, for formality, they are less entangled. Our findings suggest that for different types of styles different approaches to modeling style for style-transfer are necessary.