EditCoT: A Novel Multi-Intent Text Revision Modeling Framework

Anonymous

Abstract: Multi-intent text revision is a complex process aiming to fix all potential text defects. Inspired by Chain-of-Thought, this study introduces a multi-step edit reasoning framework (EditCoT) to model multi-intent text revision tasks using large language models (LLMs). EditCoT decomposes the text revision task into multiple rewrite reasoning steps and fixes the corresponding text defects in each reasoning step. EditCoT enhances the reasoning ability of LLMs in text editing and enables multi-intent text revision by resolving each edit intent step-by-step. We investigate the performance of EditCoT on multi-/single-intent text revision tasks. The results show that EditCoT can achieve the best performance in multi-intent text revision and present a competitive performance compared to specifically fine-tuned single-intent models. Additionally, EditCoT also exhibits good transferability to unseen edit intents

Paper Type: long

Research Area: NLP Applications

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models, Data resources, Data analysis

Languages Studied: English