Large language models show mixed reliability in time extraction tasks

Evaluating Multi-Dimensional Generalization of Large Language Models in Temporal Extraction Tasks

Computation and Language

Summary

Extracting times and events from text is tricky because language can be unclear and change depending on context. The authors studied how large language models handle these tasks across different conditions, like shifting topics or longer texts. They found models that do well on easy cases don't always perform well when things get harder or different. Using certain ways to ask the models questions helps them stay reliable across different situations. But no single test can predict how well these models will do on all types of time extraction problems.

What this means in practice

Authors

Fahmid Shahriar Iqbal, Ritam Dutt, Soumitra Das, Arnav Verma, Sagnik Ray Choudhury

Abstract

Time and event expression extraction are fundamental temporal reasoning tasks, but the problem remains difficult due to annotation ambiguity, domain sensitivity, and unstable model behavior. Existing evaluations focus on in-domain performance, offering limited insight into reliability under distribution shifts. We evaluate multiple model configurations across families, architectures, and reasoning strategies over four dimensions of generalization, examining transfer from base performance, cross-dimensional correlations, and the effects of scale, architecture, and prompting. This provides a systematic study of how prompted LLMs generalize in time and event expression extraction tasks. We find that strong base-task performance generally predicts better generalization. However, this relationship weakens under substantial distribution shifts. Inductive prompting performs most consistently across domain shift, adversarial perturbations, compositionality, and length increase, while gains from scale, architecture, and deductive and abductive prompting strategies are uneven and dimension-specific. We conclude that LLM generalization in temporal extraction tasks cannot be predicted from any single dimension alone and cannot be reliably inferred from in-domain or single-dimension evaluations, highlighting the need for reasoning strategies that generalize across dimensions.