AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts

TL;DR

AutoPrompt introduces an automated, gradient-guided method for creating prompts that effectively elicit knowledge from language models, enabling tasks like sentiment analysis and relation extraction without additional training.

Contribution

It presents AutoPrompt, a novel automated prompt generation technique that enhances the probing of language models' knowledge without finetuning or manual prompt crafting.

Findings

AutoPrompt achieves competitive performance on sentiment analysis and natural language inference.

It elicits more accurate factual knowledge than manual prompts on the LAMA benchmark.

Language models can perform relation extraction more effectively with AutoPrompt than supervised models.

Abstract

The remarkable success of pretrained language models has motivated the study of what kinds of knowledge these models learn during pretraining. Reformulating tasks as fill-in-the-blanks problems (e.g., cloze tests) is a natural approach for gauging such knowledge, however, its usage is limited by the manual effort and guesswork required to write suitable prompts. To address this, we develop AutoPrompt, an automated method to create prompts for a diverse set of tasks, based on a gradient-guided search. Using AutoPrompt, we show that masked language models (MLMs) have an inherent capability to perform sentiment analysis and natural language inference without additional parameters or finetuning, sometimes achieving performance on par with recent state-of-the-art supervised models. We also show that our prompts elicit more accurate factual knowledge from MLMs than the manually created prompts on the LAMA benchmark, and that MLMs can be used as relation extractors more effectively than supervised relation extraction models. These results demonstrate that automatically generated prompts are a viable parameter-free alternative to existing probing methods, and as pretrained LMs become more sophisticated and capable, potentially a replacement for finetuning.