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/

Zendy Plus

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

Zendy Tools

Zendy Open

Recent studies have revealed a number of pathologies of neural machinetranslation (NMT) systems. Hypotheses explaining these mostly suggest there issomething fundamentally wrong with NMT as a model or its training algorithm,maximum likelihood estimation (MLE). Most of this evidence was gathered usingmaximum a posteriori (MAP) decoding, a decision rule aimed at identifying thehighest-scoring translation, i.e. the mode. We argue that the evidencecorroborates the inadequacy of MAP decoding more than casts doubt on the modeland its training algorithm. In this work, we show that translationdistributions do reproduce various statistics of the data well, but that beamsearch strays from such statistics. We show that some of the known pathologiesand biases of NMT are due to MAP decoding and not to NMT's statisticalassumptions nor MLE. In particular, we show that the most likely translationsunder the model accumulate so little probability mass that the mode can beconsidered essentially arbitrary. We therefore advocate for the use of decisionrules that take into account the translation distribution holistically. We showthat an approximation to minimum Bayes risk decoding gives competitive resultsconfirming that NMT models do capture important aspects of translation well inexpectation.

Is MAP Decoding All You Need? The Inadequacy of the Mode in Neural Machine Translation