Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Probabilistic Formulation of Unsupervised Text Style Transfer
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 |
|---|---|---|---|
| Use the probabilistic objective with its negative-entropy regularizer instead of BT+NLL | Deep latent sequence model · Sentiment transfer (Yelp) | BLEU: 12.36 -> 18.67 | Helped |
| Use the probabilistic objective with its negative-entropy regularizer instead of BT+NLL | Deep latent sequence model · Formality transfer (GYAFC Entertainment and Music) | BLEU: 8.57 -> 18.54 | Helped |
| Use the probabilistic objective with its negative-entropy regularizer instead of BT+NLL | Deep latent sequence model · Author imitation | BLEU: 10.80 -> 10.81 | Helped |
| Use the probabilistic objective with its negative-entropy regularizer instead of UNMT | Deep latent sequence model · Related language translation (Sr-Bs) | BLEU-1: 31.4 -> 36.2 | Helped |
| Use the probabilistic objective with its negative-entropy regularizer instead of UNMT | Deep latent sequence model · Unsupervised MT (De-En) | BLEU: 32.2 -> 32.0 | Hurt |
| Add the negative-entropy regularizer to the back-translation and language-model objective (ablation) | BT+NLL · Sentiment transfer (Yelp) | BLEU: 12.36 -> 18.67 | Helped |
| Add the negative-entropy regularizer to the back-translation and language-model objective (ablation) | BT+NLL · Formality transfer (GYAFC Entertainment and Music) | BLEU: 8.57 -> 18.54 | Helped |
| Replace greedy reconstruction approximation with sample-based training (ablation) | Deep latent sequence model · Sentiment transfer (Yelp) | BLEU_r: 18.67 -> 13.34 | Hurt |
| Replace greedy reconstruction approximation with sample-based training (ablation) | Deep latent sequence model · Sentiment transfer (Yelp) | test ELBO: -2.07 -> -3.79 | Hurt |
| Propagate reconstruction gradients with straight-through Gumbel Softmax instead of stopping them (ablation) | Deep latent sequence model · Sentiment transfer (Yelp) | BLEU_r: 18.67 -> 16.17 | Hurt |
| Propagate reconstruction gradients with REINFORCE instead of stopping them (ablation) | Deep latent sequence model · Sentiment transfer (Yelp) | BLEU_r: 18.67 -> 4.08 | Hurt |
| Remove parameter sharing between transfer directions (ablation) | Deep latent sequence model · Word substitution decipherment | BLEU | Hurt |
Closest papers by meaning
- Unsupervised Text Style Transfer using Language Models as Discriminators (NeurIPS 2018 · similarity 0.91)
- Revision in Continuous Space: Unsupervised Text Style Transfer without Adversarial Learning (AAAI 2020 · similarity 0.90)
- Style Transfer in Text: Exploration and Evaluation (AAAI 2018 · similarity 0.89)
- ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer (AAAI 2024 · similarity 0.88)
- The Hyperfitting Phenomenon: Sharpening and Stabilizing LLMs for Open-Ended Text Generation (ICLR 2025 · similarity 0.85)
- Boosting Alignment for Post-Unlearning Text-to-Image Generative Models (NeurIPS 2024 · similarity 0.85)
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.