Research map / Image generation / Editing and style transfer
Content preserving text generation with attribute controls
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 proposed model instead of Cross-align | Sentiment-conditioned generation · Yelp Reviews | BLEU-1: 41.9 -> 53.0 | Helped |
| Use the proposed model instead of Cross-align | Sentiment-conditioned generation · IMDB Reviews | attribute accuracy: 88.68% -> 94.46% | Helped |
| Add adversarial loss to the reconstruction objective (ablation) | Text attribute transfer · Validation set | attribute compatibility | Helped |
| Use interpolated reconstruction loss instead of autoencoding loss alone (ablation) | Text attribute transfer · Validation set | content compatibility | Helped |
| Replace interpolated reconstruction with autoencoding plus back-translation losses (ablation) | Text attribute transfer · Validation set | attribute compatibility | Mixed |
| Use soft-sampling rather than hard-sampling during training (ablation) | Sentiment-conditioned generation · Yelp Reviews | BLEU-1: 53.0 -> 43.6 | Hurt |
| Use soft-sampling rather than hard-sampling during training (ablation) | Sentiment-conditioned generation · Yelp Reviews | attribute compatibility: 90.50% -> 92.33% | Helped |
| Add unpaired data to paired-data training and fine-tune with the proposed objective | Old English to Modern English translation · Old English to Modern English | BLEU: 10.4 -> 13.89; 11.15 -> 13.89 | Helped |
| Use unpaired data only rather than supervised paired-data training | Old English to Modern English translation · Old English to Modern English | BLEU: 10.4 -> 7.65; 11.15 -> 7.65 | Hurt |
Closest papers by meaning
- Toward Controlled Generation of Text (ICML 2017 · similarity 0.88)
- Multiple-Attribute Text Rewriting (ICLR 2019 · similarity 0.88)
- Plug and Play Language Models: A Simple Approach to Controlled Text Generation (ICLR 2020 · similarity 0.86)
- C³TG: Conflict-aware, Composite, and Collaborative Controlled Text Generation (AAAI 2026 · similarity 0.86)
- Towards Text Generation with Adversarially Learned Neural Outlines (NeurIPS 2018 · similarity 0.86)
- Successor Features for Efficient Multi-Subject Controlled Text Generation (ICML 2024 · similarity 0.83)
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.