受控文本生成技术研究综述

Review on Controlled Text Generation Technology

文本生成技术是自然语言处理的一个重要领域,利用深度学习以及语言学来帮助计算机理解文本并生成人类需要的文本。受控文本生成增加生成过程中的可控性,降低语言的毒性,提高特定领域内文本的专业性。针对样式迁移任务以及通用受控文本生成任务,首先介绍相关模型与技术,并对比这些模型技术的优缺点,其次分析相关模型技术适合应用的领域,最后提出当前受控文本生成技术的发展前景与研究方向。

Text generation technology is an important field of natural language processing,which uses deep learning and linguistics to help computers understand text and generate text that humans need.Controlled text generation increases the controllability of the generation process,reduces the toxicity of language,and improves the specialization of text in a specific domain.This paper first introduces related models and technologies for style transfer task and general controlled text generation task,compares the advantages and disadvantages of these model technologies,then analyzes the suitable application fields of related model technologies,and finally puts forward the development prospect and research direction of the current controlled text generation technology.