LaSQuE: Improved Zero-Shot Classification from Explanations Through Quantifier Modeling and Curriculum Learning

TL;DR

LaSQuE is a novel method that enhances zero-shot classification by modeling linguistic quantifiers, aggregating explanations with attention, and employing curriculum learning, leading to significant generalization improvements.

Contribution

The paper introduces LaSQuE, a new approach that leverages quantifier semantics, explanation aggregation, and curriculum learning for improved zero-shot classification from language explanations.

Findings

Up to 7% improvement in generalization to unseen tasks.

Effective modeling of quantifiers enhances explanation-based learning.

Aggregation and curriculum strategies outperform prior methods.

Abstract

A hallmark of human intelligence is the ability to learn new concepts purely from language. Several recent approaches have explored training machine learning models via natural language supervision. However, these approaches fall short in leveraging linguistic quantifiers (such as 'always' or 'rarely') and mimicking humans in compositionally learning complex tasks. Here, we present LaSQuE, a method that can learn zero-shot classifiers from language explanations by using three new strategies - (1) modeling the semantics of linguistic quantifiers in explanations (including exploiting ordinal strength relationships, such as 'always' > 'likely'), (2) aggregating information from multiple explanations using an attention-based mechanism, and (3) model training via curriculum learning. With these strategies, LaSQuE outperforms prior work, showing an absolute gain of up to 7% in generalizing to unseen real-world classification tasks.