Learning Shortcuts: On the Misleading Promise of NLU in Language Models

TL;DR

The paper investigates how shortcut learning in large language models can inflate NLU benchmark performance while harming generalization and calibration. It surveys methods to identify shortcuts and reviews data-centric and model-centric approaches to mitigate them, highlighting out-of-distribution robustness and the need for improved evaluation standards. It details the quantitative and qualitative impacts of shortcuts on performance and confidence, and discusses language variation, vocabulary shifts, and training-loss incentives as critical factors. The work emphasizes practical implications for building more robust, trustworthy NLU systems and advocates for broader, more rigorous evaluation practices in real-world settings.

Abstract

The advent of large language models (LLMs) has enabled significant performance gains in the field of natural language processing. However, recent studies have found that LLMs often resort to shortcuts when performing tasks, creating an illusion of enhanced performance while lacking generalizability in their decision rules. This phenomenon introduces challenges in accurately assessing natural language understanding in LLMs. Our paper provides a concise survey of relevant research in this area and puts forth a perspective on the implications of shortcut learning in the evaluation of language models, specifically for NLU tasks. This paper urges more research efforts to be put towards deepening our comprehension of shortcut learning, contributing to the development of more robust language models, and raising the standards of NLU evaluation in real-world scenarios.