A Mandarin-Cantonese Parallel Corpus with Formality Ranking

Formality-controlled machine translation allows users to specify the formality level of the target sentence, so that it would be suitable for the intended audience. While formality-annotated datasets have been constructed for some major languages, no such resource is currently available for Cantonese. This paper presents a Mandarin-Cantonese parallel corpus with 300 Mandarin sentences, each of which is aligned to a list of five or more Cantonese sentences ranked according to their level of formality. To our knowledge, this is the first parallel translation corpus with manual formality ranking, which provides more nuanced judgment than the formal/informal dichotomy in most current formality-annotated datasets. This corpus can support future research towards more fine-grained notions of formality in terminology, translation and text style transfer. © 2025 Copyright for this paper by its authors.