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

Solid evaluation of neural machine translation (NMT) is key to itsunderstanding and improvement. Current evaluation of an NMT system is usuallybuilt upon a heuristic decoding algorithm (e.g., beam search) and an evaluationmetric assessing similarity between the translation and golden reference.However, this system-level evaluation framework is limited by evaluating onlyone best hypothesis and search errors brought by heuristic decoding algorithms.To better understand NMT models, we propose a novel evaluation protocol, whichdefines model errors with model's ranking capability over hypothesis space. Totackle the problem of exponentially large space, we propose two approximationmethods, top region evaluation along with an exact top-$k$ decoding algorithm,which finds top-ranked hypotheses in the whole hypothesis space, and MonteCarlo sampling evaluation, which simulates hypothesis space from a broaderperspective. To quantify errors, we define our NMT model errors by measuringdistance between the hypothesis array ranked by the model and the ideallyranked hypothesis array. After confirming the strong correlation with humanjudgment, we apply our evaluation to various NMT benchmarks and modelarchitectures. We show that the state-of-the-art Transformer models faceserious ranking issues and only perform at the random chance level in the topregion. We further analyze model errors on architectures with different depthsand widths, as well as different data-augmentation techniques, showing howthese factors affect model errors. Finally, we connect model errors with thesearch algorithms and provide interesting findings of beam search inductivebias and correlation with Minimum Bayes Risk (MBR) decoding.

Digging Errors in NMT: Evaluating and Understanding Model Errors from Partial Hypothesis Space