Abstract
Generic text rewriting is a prevalent large language model (LLM) application that covers diverse real-world tasks, such as style transfer, fact correction, and email editing. These tasks vary in rewriting objectives (e.g., factual consistency vs. semantic preservation), making it challenging to develop a unified model that excels across all dimensions. Existing methods often specialize in either a single task or a specific objective, limiting their generalizability. In this work, we introduce a generic model proficient in factuality, stylistic, and conversational rewriting tasks. To simulate real-world user rewrite requests, we construct a conversational rewrite dataset, ChatRewrite, that presents ``natural''-sounding instructions, from raw emails using LLMs. Combined with other popular rewrite datasets, including LongFact for the factuality rewrite task and RewriteLM for the stylistic rewrite task, this forms a broad benchmark for training and evaluating generic rewrite models. To align with task-specific objectives, we propose Dr Genre, a Decoupled-reward learning framework for Generic rewriting, that utilizes objective-oriented reward models with a task-specific weighting. Evaluation shows that \approach delivers higher-quality rewrites across all targeted tasks, improving objectives including instruction following (agreement), internal consistency (coherence), and minimal unnecessary edits (conciseness).
Abstract (translated)
通用文本重写是一项广泛应用于大型语言模型（LLM）的实际任务，涵盖了多种现实世界的应用场景，例如文体转换、事实校正以及电子邮件编辑。这些任务在重写目标方面各有不同（如：事实一致性 vs 语义保持），这使得开发一个能在所有维度上表现优秀的统一模型变得具有挑战性。现有的方法通常专注于单一任务或特定目标，从而限制了它们的通用性。在这项工作中，我们引入了一个擅长于事实性、风格性和对话式重写任务的通用模型。为了模拟现实世界中用户对于文本重写的请求，我们构建了一个基于对话式的重写数据集ChatRewrite，该数据集提供了使用LLM处理原始邮件时“自然”听起来的指令。结合LongFact（用于事实一致性重写任务）和RewriteLM（用于风格性重写任务）等其他流行的数据集，形成了一个广泛的基准测试环境以训练和评估通用文本重写模型。 为了使模型更符合特定的任务目标，我们提出了Dr Genre（Decoupled-reward learning framework for Generic rewriting），即解耦奖励学习框架。此框架使用面向具体目标的奖励模型，并结合了任务特异性的加权策略。评估结果表明，\approach 在所有被关注的任务中提供了更高质量的重写文本，提升了包括指令遵循（一致性）、内部连贯性（语义一致）和最小化不必要的修改（简洁度）在内的多个指标表现。
URL
https://arxiv.org/abs/2503.06781
