Are Prompt-based Models Clueless?

TL;DR

This paper investigates whether few-shot prompt-based language models exploit superficial dataset cues, finding that they do, especially on instances with superficial cues, raising questions about their true understanding capabilities.

Contribution

It provides an empirical analysis showing that prompt-based models also exploit superficial cues, challenging assumptions about their robustness and understanding.

Findings

Prompt-based models exploit superficial cues in datasets.

Models perform poorly on instances without superficial cues.

Prompting does not fully eliminate reliance on superficial cues.

Abstract

Finetuning large pre-trained language models with a task-specific head has advanced the state-of-the-art on many natural language understanding benchmarks. However, models with a task-specific head require a lot of training data, making them susceptible to learning and exploiting dataset-specific superficial cues that do not generalize to other datasets. Prompting has reduced the data requirement by reusing the language model head and formatting the task input to match the pre-training objective. Therefore, it is expected that few-shot prompt-based models do not exploit superficial cues. This paper presents an empirical examination of whether few-shot prompt-based models also exploit superficial cues. Analyzing few-shot prompt-based models on MNLI, SNLI, HANS, and COPA has revealed that prompt-based models also exploit superficial cues. While the models perform well on instances with superficial cues, they often underperform or only marginally outperform random accuracy on instances without superficial cues.