Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Text Style Transfer using Language Models as Discriminators
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace a binary discriminator with a target-domain language model | unsupervised text style transfer · Word substitution decipherment, 40% | BLEU: 77.1 -> 80.0 | Helped |
| replace a binary discriminator with a target-domain language model | unsupervised text style transfer · Related language translation, cn--tw | BLEU: 60.1 -> 81.6 | Helped |
| add adversarial negative-sample training to the language model (ablation) | LM decipherment · Word substitution decipherment | BLEU: 89.0 -> 89.1; 49.3 -> 44.2 | Mixed |
| add a classifier loss to the language-model discriminator (ablation) | LM sentiment transfer · Sentiment modification (negative to positive) | Accuracy: 83.3 -> 91.2 | Helped |
| add a classifier loss to the language-model discriminator (ablation) | LM sentiment transfer · Sentiment modification (negative to positive) | BLEU: 38.6 -> 57.8 | Hurt |
| add a classifier loss to the language-model discriminator (ablation) | LM sentiment transfer · Sentiment modification, 500 human-annotated sentences | BLEU: 13.4 -> 22.3 | Helped |
| use target-domain language-model feedback rather than the comparable CNN discriminator | unsupervised sentiment transfer · Sentiment modification (negative to positive) | PPL_X: 50.4 -> 30.3 | Helped |
Closest papers by meaning
- A Probabilistic Formulation of Unsupervised Text Style Transfer (ICLR 2020 · similarity 0.91)
- Style Transfer in Text: Exploration and Evaluation (AAAI 2018 · similarity 0.89)
- Revision in Continuous Space: Unsupervised Text Style Transfer without Adversarial Learning (AAAI 2020 · similarity 0.88)
- Comparing Unsupervised Word Translation Methods Step by Step (NeurIPS 2019 · similarity 0.86)
- Branch-GAN: Improving Text Generation with (not so) Large Language Models (ICLR 2024 · similarity 0.86)
- Improving Neural Language Modeling via Adversarial Training (ICML 2019 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.