2018/08/13 by Vineet John, Lili Mou, John, Vineet +5 · 7 citations
Computer Science · #68T50 #Computation and Language (cs.CL) #FOS: Computer and information sciences #I.2.7 #Natural Language Processing Techniques #Speech Recognition and Synthesis #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1808.04339
openalex publication_date 2018/08/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
This paper tackles the problem of disentangling the latent variables of style\nand content in language models. We propose a simple yet effective approach,\nwhich incorporates auxiliary multi-task and adversarial objectives, for label\nprediction and bag-of-words prediction, respectively. We show, both\nqualitatively and quantitatively, that the style and content are indeed\ndisentangled in the latent space. This disentangled latent representation\nlearning method is applied to style transfer on non-parallel corpora. We\nachieve substantially better results in terms of transfer accuracy, content\npreservation and language fluency, in comparison to previous state-of-the-art\napproaches.\n