Elaboration-Generating Commonsense Question Answering at Scale

TL;DR

This paper introduces a scalable method for commonsense question answering by fine-tuning small language models to generate background elaborations, improving performance while significantly reducing computational costs.

Contribution

The work presents a novel framework that alternates between an elaboration generator and answer predictor, outperforming similar-sized models and approaching GPT-3's performance.

Findings

Outperforms models with similar size on four benchmarks

Uses less than 0.5% of GPT-3's parameters

High-quality elaborations confirmed by human evaluation

Abstract

In question answering requiring common sense, language models (e.g., GPT-3) have been used to generate text expressing background knowledge that helps improve performance. Yet the cost of working with such models is very high; in this work, we finetune smaller language models to generate useful intermediate context, referred to here as elaborations. Our framework alternates between updating two language models -- an elaboration generator and an answer predictor -- allowing each to influence the other. Using less than 0.5% of the parameters of GPT-3, our model outperforms alternatives with similar sizes and closes the gap on GPT-3 on four commonsense question answering benchmarks. Human evaluations show that the quality of the generated elaborations is high.