Salvato in:
Dettagli Bibliografici
Autori principali: Li, Yufei, Nham, John, Jawahar, Ganesh, Shu, Lei, Uthus, David, Sung, Yun-Hsuan, Yang, Chengrun, Rolnick, Itai, Qiao, Yi, Liu, Cong
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2503.06781
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915189535801344
author Li, Yufei
Nham, John
Jawahar, Ganesh
Shu, Lei
Uthus, David
Sung, Yun-Hsuan
Yang, Chengrun
Rolnick, Itai
Qiao, Yi
Liu, Cong
author_facet Li, Yufei
Nham, John
Jawahar, Ganesh
Shu, Lei
Uthus, David
Sung, Yun-Hsuan
Yang, Chengrun
Rolnick, Itai
Qiao, Yi
Liu, Cong
contents 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).
format Preprint
id arxiv_https___arxiv_org_abs_2503_06781
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dr Genre: Reinforcement Learning from Decoupled LLM Feedback for Generic Text Rewriting
Li, Yufei
Nham, John
Jawahar, Ganesh
Shu, Lei
Uthus, David
Sung, Yun-Hsuan
Yang, Chengrun
Rolnick, Itai
Qiao, Yi
Liu, Cong
Computation and Language
Artificial Intelligence
Machine Learning
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).
title Dr Genre: Reinforcement Learning from Decoupled LLM Feedback for Generic Text Rewriting
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2503.06781