Most deep learning work on TST adopts a data-driven definition of style. The data-driven definition of style is different from the linguistic or rule-based definition of style, which theoretically constrains what constitutes a style and what not, such as a style guide that requires that formal text not include any contraction, e.g., “isn’t.”

Data-Driven Definition of Style

The advantage of the data-driven style is that it can marry well with deep learning methods because most neural models learn the concept of style by learning to distinguish the multiple style corpora. For the (non-data-driven) linguistic style, although it is under-explored in the existing deep learning works of TST, there's a discussion of how potential future works can learn TST of linguistics styles with no matched data.

Comparison between Data-driven Style and linguistic Style

Linguistic Definition of Style: an intuitive notion of style refers to the manner in which the semantics is expressed. Just as everyone has their own signatures, style originates as the characteristics inherent to every person’s utterance, which can be expressed through the use of certain stylistic devices such as
metaphors, as well as choice of words, syntactic structures, and so on. Style can also go beyond the sentence level to the discourse level, such as the stylistic structure of the entire piece of the work, for example, stream of consciousness, or flashbacks.

Hamilton College

The style transfer tasks mainly discuss task formulation and introducing the notations that will be used. It also lists all the common subtasks for neural TST which can save the literature review efforts for future researchers.

Style Transfer Tasks

Jin, Di, et al. “Deep Learning for Text Style Transfer: A Survey.” ACL Anthology, https://aclanthology.org/2022.cl-1.6/. 

Learn Before

Related

Learn After