English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

While non-autoregressive (NAR) models are showing great promise for machinetranslation, their use is limited by their dependence on knowledge distillationfrom autoregressive models. To address this issue, we seek to understand whydistillation is so effective. Prior work suggests that distilled training datais less complex than manual translations. Based on experiments with theLevenshtein Transformer and the Mask-Predict NAR models on the WMT14German-English task, this paper shows that different types of complexity havedifferent impacts: while reducing lexical diversity and decreasing reorderingcomplexity both help NAR learn better alignment between source and target, andthus improve translation quality, lexical diversity is the main reason whydistillation increases model confidence, which affects the calibration ofdifferent NAR models differently.

How Does Distilled Data Complexity Impact the Quality and Confidence of Non-Autoregressive Machine Translation?