AI Paper Atlas

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

AAAI 2019 · Architectures and efficiency › RNNs and normalization · approach: Non-autoregressive translation

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 changedWhereMeasuredEffect
add heredity (decoder input embeddings) to non-autoregressive model (ablation)+NAT · Switchboard testF-score: 87.7 -> 89.0Improved
add deletion-penalty loss (lambda 1.5) (ablation)T-TDD · Switchboard testF-score: 87.8 -> 88.0Improved
increase lambda from 1.5 to 2 (ablation)T-TDD · Switchboard testF-score: 88.0 -> 87.5Hurt
set alpha 0.05 vs 0.20 (ablation)T-TDD · Switchboard testF-score: 85.3 -> 87.8Improved
increase alpha from 0.20 to 0.30 (ablation)T-TDD · Switchboard testF-score: 87.8 -> 87.5Hurt
add 300k artificial DAE data (ablation)T-TDD · Switchboard testF-score: 88.0 -> 88.7Improved
add 50k artificial DAE data (ablation)T-TDD · Switchboard testF-score: 88.0 -> 88.1No clear effect
mix artificial data directly into real training data instead of DAE initialization (ablation)T-TDD · SwitchboardF-scoreHurt
more artificial data beyond 300k (ablation)T-TDD · SwitchboardF-scoreNo clear effect
replace autoregressive decoding with non-autoregressive +NATT-TDD · Switchboarddecoding speed: 0.0012s/token -> 0.0002s/tokenImproved
non-autoregressive +NAT training costT-TDD · Switchboardtrain time: 12h -> 20hHurt
add NAT to translation-based TDD on non-repetitionsTransition-based · Switchboard testF-score: 68.7 -> 78.3Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy