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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Unsupervised rationale extraction aims to extract concise and contiguous textsnippets to support model predictions without any annotated rationale. Previousstudies have used a two-phase framework known as the Rationalizing NeuralPrediction (RNP) framework, which follows a generate-then-predict paradigm.They assumed that the extracted explanation, called rationale, should besufficient to predict the golden label. However, the assumption above deviatesfrom the original definition and is too strict to perform well. Furthermore,these two-phase models suffer from the interlocking problem and spuriouscorrelations. To solve the above problems, we propose a novel single-phaseframework called You Only Forward Once (YOFO), derived from a relaxed versionof rationale where rationales aim to support model predictions rather than makepredictions. In our framework, A pre-trained language model like BERT isdeployed to simultaneously perform prediction and rationalization with lessimpact from interlocking or spurious correlations. Directly choosing theimportant tokens in an unsupervised manner is intractable. Instead of directlychoosing the important tokens, YOFO gradually removes unimportant tokens duringforward propagation. Through experiments on the BeerAdvocate and Hotel Reviewdatasets, we demonstrate that our model is able to extract rationales and makepredictions more accurately compared to RNP-based models. We observe animprovement of up to 18.4\% in token-level F1 compared to previousstate-of-the-art methods. We also conducted analyses and experiments to explorethe extracted rationales and token decay strategies. The results show that YOFOcan extract precise and important rationales while removing unimportant tokensin the middle part of the model.

You Only Forward Once: Prediction and Rationalization in A Single Forward Pass