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

Rationalization empowers deep learning models with self-explainingcapabilities through a cooperative game, where a generator selects asemantically consistent subset of the input as a rationale, and a subsequentpredictor makes predictions based on the selected rationale. In this paper, wediscover that rationalization is prone to a problem named \emph{rationaleshift}, which arises from the algorithmic bias of the cooperative game.Rationale shift refers to a situation where the semantics of the selectedrationale may deviate from the original input, but the predictor still producesaccurate predictions based on the deviation, resulting in a compromisedgenerator with misleading feedback.  To address this issue, we first demonstrate the importance of the alignmentbetween the rationale and the full input through both empirical observationsand theoretical analysis. Subsequently, we introduce a novel approach calledDAR (\textbf{D}iscriminatively \textbf{A}ligned \textbf{R}ationalization),which utilizes an auxiliary module pretrained on the full input todiscriminatively align the selected rationale and the original input. Wetheoretically illustrate how DAR accomplishes the desired alignment, therebyovercoming the rationale shift problem. The experiments on two widely usedreal-world benchmarks show that the proposed method significantly improves theexplanation quality (measured by the overlap between the model-selectedexplanation and the human-annotated rationale) as compared to state-of-the-arttechniques. Additionally, results on two synthetic settings further validatethe effectiveness of DAR in addressing the rationale shift problem.

Enhancing the Rationale-Input Alignment for Self-explaining Rationalization