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

In Neural Machine Translation (and, more generally, conditional languagemodeling), the generation of a target token is influenced by two types ofcontext: the source and the prefix of the target sequence. While many attemptsto understand the internal workings of NMT models have been made, none of themexplicitly evaluates relative source and target contributions to a generationdecision. We argue that this relative contribution can be evaluated by adoptinga variant of Layerwise Relevance Propagation (LRP). Its underlying'conservation principle' makes relevance propagation unique: differently fromother methods, it evaluates not an abstract quantity reflecting tokenimportance, but the proportion of each token's influence. We extend LRP to theTransformer and conduct an analysis of NMT models which explicitly evaluatesthe source and target relative contributions to the generation process. Weanalyze changes in these contributions when conditioning on different types ofprefixes, when varying the training objective or the amount of training data,and during the training process. We find that models trained with more datatend to rely on source information more and to have more sharp tokencontributions; the training process is non-monotonic with several stages ofdifferent nature.

Analyzing the Source and Target Contributions to Predictions in Neural Machine Translation