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 is fundamental to human reasoning and learning. NLP modelstrained to produce rationales along with predictions, calledself-rationalization models, have been investigated for their interpretabilityand utility to end-users. However, the extent to which training withhuman-written rationales facilitates learning remains an under-exploredquestion. We ask whether training models to self-rationalize can aid in theirlearning to solve tasks for the right reasons. Specifically, we evaluate howtraining self-rationalization models with free-text rationales affectsrobustness to spurious correlations in fine-tuned encoder-decoder anddecoder-only models of six different sizes. We evaluate robustness to spuriouscorrelations by measuring performance on 1) manually annotated challengedatasets and 2) subsets of original test sets where reliance on spuriouscorrelations would fail to produce correct answers. We find that whileself-rationalization can improve robustness to spurious correlations inlow-resource settings, it tends to hurt robustness in higher-resource settings.Furthermore, these effects depend on model family and size, as well as onrationale content. Together, our results suggest that explainability can comeat the cost of robustness; thus, appropriate care should be taken when trainingself-rationalizing models with the goal of creating more trustworthy models.

Does Self-Rationalization Improve Robustness to Spurious Correlations?