He Said, She Said: Style Transfer for Shifting the Perspective of Dialogues

TL;DR

This paper introduces a new perspective shift style transfer task for dialogues, transforming informal first-person conversations into formal third-person rephrasing, with applications improving dialogue summarization and extractive summarization models.

Contribution

It defines the perspective shift task for dialogues, explores baseline approaches, and demonstrates its benefits for dialogue summarization and extractive model training.

Findings

Perspective shifting improves zero-shot summarization performance.

Supervised models trained on shifted data outperform those trained on original dialogues.

The approach addresses coreference, emotion attribution, and informal text interpretation.

Abstract

In this work, we define a new style transfer task: perspective shift, which reframes a dialogue from informal first person to a formal third person rephrasing of the text. This task requires challenging coreference resolution, emotion attribution, and interpretation of informal text. We explore several baseline approaches and discuss further directions on this task when applied to short dialogues. As a sample application, we demonstrate that applying perspective shifting to a dialogue summarization dataset (SAMSum) substantially improves the zero-shot performance of extractive news summarization models on this data. Additionally, supervised extractive models perform better when trained on perspective shifted data than on the original dialogues. We release our code publicly.