Finding Sense in Nonsense with Generated Contexts: Perspectives from Humans and Language Models

In three linesComparative study on humans' and LLMs' ability to distinguish anomalous sentences from truly nonsensical ones. Analysis of five semantically deviant datasets with and without context. Finding: most sentences rated as anomalous can be interpreted with context; LLMs effectively generate plausible contexts.

Summary generated by Claude — human-verified