Toward Zero-Shot Instruction Following

TL;DR

This paper introduces a more realistic zero-shot instruction following setting that leverages task definitions and proposes strategies to improve model understanding, achieving state-of-the-art results on Super-NaturalInstructions.

Contribution

It presents a novel zero-shot instruction following framework using critical sentence identification and ranking objectives to enhance task comprehension from definitions.

Findings

Achieved state-of-the-art performance on Super-NaturalInstructions

Effective identification of critical sentences improves task understanding

Ranking objectives enhance model output accuracy

Abstract

This work proposes a challenging yet more realistic setting for zero-shot cross-task generalization: zero-shot instruction following, presuming the existence of a paragraph-style task definition while no demonstrations exist. To better learn the task supervision from the definition, we propose two strategies: first, to automatically find out the critical sentences in the definition; second, a ranking objective to force the model to generate the gold outputs with higher probabilities when those critical parts are highlighted in the definition. The joint efforts of the two strategies yield state-of-the-art performance on the Super-NaturalInstructions. Our code is available on GitHub.