AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Multi-Task Self-Supervised Learning for Disfluency Detection

Shaolei Wang, Wangxiang Che, Qi Liu, Pengda Qin, Ting Liu, William Yang Wang

AAAI 2020 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Self-supervised speech and brain decoding · 48 citations (OpenAlex)

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
combine tagging and sentence-classification pre-training rather than using either task alone (ablation)Our self-supervised model · Switchboard test, FullF1: 87.7 / 87.0 -> 90.2Improved
combine tagging and sentence-classification pre-training rather than using either task alone (ablation)Our self-supervised model · Switchboard test, 1000 sentsF1: 82.3 / 80.7 -> 86.1Improved
add self-supervised pre-training instead of random initialization (ablation)Transformer-based disfluency detector · Switchboard test, FullF1: 78.1 -> 90.2Improved
add self-supervised pre-training instead of random initialization (ablation)Transformer-based disfluency detector · Switchboard test, 1000 sentsF1: 62.1 -> 86.1Improved
increase model size across the tested configurations (ablation)Our self-supervised model · Switchboard dev, 1000 sentsF1: 78.1 -> 79.7 -> 81.8 -> 82.9 -> 84.5 -> 86.3Improved
increase model size across the tested configurations (ablation)Our self-supervised model · Switchboard dev, FullF1: 85.7 -> 86.8 -> 88.2 -> 89.1 -> 89.6 -> 90.3Improved
concatenate representations from BERT and the self-supervised modelSwitchboard disfluency detector · Switchboard test, FullF1: 90.2 / 90.1 -> 91.4Improved
concatenate representations from BERT and the self-supervised modelSwitchboard disfluency detector · Switchboard test, 1000 sentsF1: 86.1 / 82.4 -> 87.8Improved
pre-train on increasing amounts of automatically generated pseudo dataOur self-supervised model · SwitchboardF1Improved
directly merge pseudo disfluent sentences with gold training data instead of using the proposed pre-training processTransition-based and transformer-based baselines · SwitchboardF1Hurt

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