AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

A Survey on Non-Autoregressive Generation for Neural Machine Translation and Beyond

Yisheng Xiao, Lijun Wu, Junliang Guo, Juntao Li, Min Zhang, Tao Qin, Tie‐Yan Liu

TPAMI 2023 · Architectures and efficiency › RNNs and normalization · approach: Non-autoregressive translation · 118 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
add correction and reveal-position to CMLMCMLM · WMT14 EN-DEBLEU: 27.03* -> 28.37*Improved
use Viterbi decoding instead of default decodingDA-Transformer · WMT14 EN-DEBLEU: 27.49 -> 26.89Hurt
add fuzzy alignments to DA-TransformerDA-Transformer · WMT14 EN-DEBLEU: 27.49 -> 27.53Improved
add glancing training to NATNAT · WMT14 EN-DEBLEU: 17.69 -> 25.21Improved
add iterative refinement to NATNAT · WMT14 EN-DEBLEU: 17.69 -> 21.61Improved
increase decoding iterations (iterative vs one-step)NAT · WMT14 EN-DEdecoding speedup: 15.6x -> 1.5xHurt
add NAT drafting with AT verificationsemi-autoregressive decoding · WMT14 EN-DEBLEU: 26.90 -> 28.89*Improved
add sequence-level knowledge distillation (GLAT vs CMLM-like KD models)one-step NAT · WMT14 EN-DEBLEU: 25.21 -> 27.03*Mixed
add CTC loss with order-agnostic lossorder-agnostic cross-entropy · WMT14 EN-DEBLEU: 26.10* -> 27.41Improved
add n-gram level order-agnostic lossorder-agnostic cross-entropy · WMT14 EN-DEBLEU: 26.10* -> 26.50*Improved
add pre-trained XLM-R with decoratorNAT · WMT14 EN-DEBLEU: 27.49 -> 29.80Improved
add deep supervision layer-wise predictionsone-step NAT · WMT16 EN-ROBLEU: 31.19 -> 34.17Improved

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