2019/08/25 by Akhilesh Sudhakar, Sudhakar, Akhilesh, Bhargav Upadhyay +3 · 1 citation
Computer Science · Social Sciences · #Computation and Language (cs.CL) #Computational and Text Analysis Methods #FOS: Computer and information sciences #Machine Learning (cs.LG) #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1908.09368
openalex publication_date 2019/08/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Text style transfer is the task of transferring the style of text having\ncertain stylistic attributes, while preserving non-stylistic or content\ninformation. In this work we introduce the Generative Style Transformer (GST) -\na new approach to rewriting sentences to a target style in the absence of\nparallel style corpora. GST leverages the power of both, large unsupervised\npre-trained language models as well as the Transformer. GST is a part of a\nlarger `Delete Retrieve Generate' framework, in which we also propose a novel\nmethod of deleting style attributes from the source sentence by exploiting the\ninner workings of the Transformer. Our models outperform state-of-art systems\nacross 5 datasets on sentiment, gender and political slant transfer. We also\npropose the use of the GLEU metric as an automatic metric of evaluation of\nstyle transfer, which we found to compare better with human ratings than the\npredominantly used BLEU score.\n