Research map / Architectures and efficiency / RNNs and normalization
Adapting Translation Models for Transcript Disfluency Detection
Qianqian Dong, Feng Wang, Zhen Yang, Wei Chen, Shuang Xu, Bo Xu
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 heredity (decoder input embeddings) to non-autoregressive model (ablation) | +NAT · Switchboard test | F-score: 87.7 -> 89.0 | Improved |
| add deletion-penalty loss (lambda 1.5) (ablation) | T-TDD · Switchboard test | F-score: 87.8 -> 88.0 | Improved |
| increase lambda from 1.5 to 2 (ablation) | T-TDD · Switchboard test | F-score: 88.0 -> 87.5 | Hurt |
| set alpha 0.05 vs 0.20 (ablation) | T-TDD · Switchboard test | F-score: 85.3 -> 87.8 | Improved |
| increase alpha from 0.20 to 0.30 (ablation) | T-TDD · Switchboard test | F-score: 87.8 -> 87.5 | Hurt |
| add 300k artificial DAE data (ablation) | T-TDD · Switchboard test | F-score: 88.0 -> 88.7 | Improved |
| add 50k artificial DAE data (ablation) | T-TDD · Switchboard test | F-score: 88.0 -> 88.1 | No clear effect |
| mix artificial data directly into real training data instead of DAE initialization (ablation) | T-TDD · Switchboard | F-score | Hurt |
| more artificial data beyond 300k (ablation) | T-TDD · Switchboard | F-score | No clear effect |
| replace autoregressive decoding with non-autoregressive +NAT | T-TDD · Switchboard | decoding speed: 0.0012s/token -> 0.0002s/token | Improved |
| non-autoregressive +NAT training cost | T-TDD · Switchboard | train time: 12h -> 20h | Hurt |
| add NAT to translation-based TDD on non-repetitions | Transition-based · Switchboard test | F-score: 68.7 -> 78.3 | Improved |
Closest papers by meaning
- DiffNorm: Self-Supervised Normalization for Non-autoregressive Speech-to-speech Translation (NeurIPS 2024 · similarity 0.87)
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 2025 · similarity 0.85)
- Enhancing Large Language Model Performance with Gradient-Based Parameter Selection (AAAI 2025 · similarity 0.84)
- BLoB: Bayesian Low-Rank Adaptation by Backpropagation for Large Language Models (NeurIPS 2024 · similarity 0.84)
- Joint Fine-tuning and Conversion of Pretrained Speech and Language Models towards Linear Complexity (ICLR 2025 · similarity 0.84)
- DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors (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.
