Most stories present events in the order they happen, but nonlinear narratives deliberately break that sequence. A new paper on arXiv introduces a benchmark designed to test whether language models can handle this kind of storytelling.

According to the abstract, the authors built a corpus and benchmark from 83 published works. The benchmark measures whether language models can produce narratives where the order of telling differs from the order of events. The abstract does not report results, so the paper's findings are not yet available from the announcement.

The work highlights a distinction between chronological and narrative order, a structural challenge that matters for creative writing and for evaluating how models reason about story form.