TST-GAN: A Legal Document Generation Model Based on Text Style Transfer
Xiaolin Li, Lei Huang, Yifan Zhou, Changcheng Shao · 2021
Legal documents need to be written according to certain rules and grammar. With the development of deep learning in the field of natural language processing, the human cost of legal personnel in writing legal documents has been greatly reduced. Text generation has always been a research topic in the field of natural language processing. In recent years, the emergence of generation confrontation network provides new ideas and achievements for text generation. The traditional generative countermeasure network model is to generate text sequence by inputting keywords into the generator, and then judge the difference between the generated text and the real text by discriminator. However, it takes a certain amount of human resources to mark the keywords in each article, and the training of the generator is difficult to achieve good results due to the lack of good parallel corpus. Based on the generation of confrontation model, this paper proposes a text style transfer model tst-gan (general adverse networks based on text style transfer) for legal documents. First, the discriminator is used to train the real text, and then the generator generates the original text into the format of the target text by the way of text style transfer. The experimental results show that the model can reduce the workload and generate ideal legal texts.