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

Neural language models (LMs) have achieved impressive results on variouslanguage-based reasoning tasks by utilizing latent knowledge encoded in theirown pretrained parameters. To make this reasoning process more explicit, recentworks retrieve a rationalizing LM's internal knowledge by training or promptingit to generate free-text rationales, which can be used to guide taskpredictions made by either the same LM or a separate reasoning LM. However,rationalizing LMs require expensive rationale annotation and/or computation,without any assurance that their generated rationales improve LM taskperformance or faithfully reflect LM decision-making. In this paper, we proposePINTO, an LM pipeline that rationalizes via prompt-based learning, and learnsto faithfully reason over rationales via counterfactual regularization. First,PINTO maps out a suitable reasoning process for the task input by prompting afrozen rationalizing LM to generate a free-text rationale. Second, PINTO'sreasoning LM is fine-tuned to solve the task using the generated rationale ascontext, while regularized to output less confident predictions when therationale is perturbed. Across four datasets, we show that PINTO significantlyimproves the generalization ability of the reasoning LM, yielding higherperformance on both in-distribution and out-of-distribution test sets. Also, wefind that PINTO's rationales are more faithful to its task predictions thanthose generated by competitive baselines.

PINTO: Faithful Language Reasoning Using Prompt-Generated Rationales