Conventional text style transfer approaches for natural language focus on sentence-level style transfer without considering contextual information, and the style is described with attributes (e.g., formality). When applying style transfer on conversations such as task-oriented dialogues, existing approaches suffer from these limitations as context can play an important role and the style attributes are often difficult to define in conversations. In this paper, we introduce conversation style transfer as a few-shot learning problem, where the model learns to perform style transfer by observing only the target-style dialogue examples. We propose a novel in-context learning approach to solve the task with style-free dialogues as a pivot. Human evaluation shows that by incorporating multi-turn context, the model is able to match the target style while having better appropriateness and semantic correctness compared to utterance-level style transfer. Additionally, we show that conversation style transfer can also benefit downstream tasks. Results on multi-domain intent classification tasks show improvement in F1 scores after transferring the style of training data to match the style of test data.

翻译：传统自然语言文本风格迁移方法聚焦于句子级别的风格转换，未考虑上下文信息，且风格通过属性（如正式性）进行描述。将风格迁移应用于任务型对话等对话场景时，现有方法存在局限性——上下文可能发挥重要作用，且对话中的风格属性往往难以明确定义。本文提出将对话风格迁移作为少样本学习问题，模型仅通过观察目标风格的对话示例即可学习执行风格迁移。我们创新性地提出一种上下文内学习方法，以无风格对话为枢轴解决该任务。人工评估表明，通过融入多轮上下文，模型既能匹配目标风格，又在适切性和语义正确性上优于话语级风格迁移。此外，我们证明对话风格迁移对下游任务亦有裨益。在多领域意图分类任务中，将训练数据风格迁移至测试数据风格后，F1分数获得显著提升。