Elaboration-Generating Commonsense Question Answering at Scale
This work addresses the cost barrier for deploying commonsense reasoning in AI applications, offering a more efficient solution that is incremental in improving model efficiency.
The paper tackles the high cost of using large language models like GPT-3 for commonsense question answering by finetuning smaller models to generate intermediate context called elaborations, achieving performance that outperforms similar-sized alternatives and narrows the gap with GPT-3 on four benchmarks using less than 0.5% of GPT-3's parameters.
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.