Research map / Large language models / Language, safety and interpretability
Unsupervised Controllable Text Formalization
Parag Jain, Abhijit Mishra, Amar Prakash Azad, Karthik Sankaranarayanan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add the control predictor (ablation) | CtrlWithPredictor · Formalization test set | LM Score: 0.32 -> 0.34 | Improved |
| add the control predictor (ablation) | CtrlWithPredictor · Formalization test set | Readability: 0.538 -> 0.568 | Improved |
| add the control predictor (ablation) | CtrlWithPredictor · Formalization test set | Relatedness: 0.77 -> 0.72 | Hurt |
| replace one-round exploration-exploitation with iterative cycles (ablation) | CtrlWithPredictor · Formalization test set | Readability: 0.554 -> 0.568 | Improved |
| replace one-round exploration-exploitation with iterative cycles (ablation) | CtrlWithPredictor · Formalization test set | LM Score: 0.30 -> 0.34 | Improved |
| replace one-round exploration-exploitation with iterative cycles (ablation) | CtrlWithPredictor · Formalization test set | Relatedness: 0.78 -> 0.72 | Hurt |
| increase the formalization control from Mid to High | Formalness-High · Reversed simplification test set | Readability: 0.52 -> 0.74 | Improved |
| use the unsupervised method instead of Seq2Seq | Formalness-Mid · Reversed simplification test set | BLEU: 38.37 -> 21.81 | Hurt |
| use the unsupervised method instead of Mueller et al. | Formalness-Mid · Reversed simplification test set | Relatedness: 0.41 -> 0.58 | Improved |
| use the unsupervised method instead of Mueller et al. | Formalness-Mid · Reversed simplification test set | BLEU: 5.09 -> 21.81 | Improved |
Closest papers by meaning
- On Variational Learning of Controllable Representations for Text without Supervision (ICML 2020 · similarity 0.86)
- Controllable Text Generation with Neurally-Decomposed Oracle (NeurIPS 2022 · similarity 0.86)
- Controlled Text Generation with Natural Language Instructions (ICML 2023 · similarity 0.85)
- Attention Biasing and Context Augmentation for Zero-Shot Control of Encoder-Decoder Transformers for Natural Language Generation (AAAI 2022 · similarity 0.85)
- Controlled Text Generation as Continuous Optimization with Multiple Constraints (NeurIPS 2021 · similarity 0.85)
- Hierarchical Autoregressive Transformers: Combining Byte- and Word-Level Processing for Robust, Adaptable Language Models (ICLR 2025 · similarity 0.84)
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.
