Temporal Reasoning in the Era of LLMs: A Survey

ACL ARR 2025 May Submission7545 Authors

Abstract: Temporal reasoning is a critical component of natural language understanding, yet it remains a challenging task due to the inherent ambiguity and implicit nature of temporal information in language. The rise of large language models (LLMs) has sparked interest in assessing their ability to reason about time. However, existing research adopts diverse methodologies, proposing different tasks, benchmarks, and evaluation strategies, making it difficult to form a cohesive view of the field. In this survey, we provide a comprehensive overview of recent work on temporal reasoning in the context of LLMs. We examine the range of tasks, benchmarks, and fine-tuning approaches, and compare these with pre-LLM temporal reasoning tasks. Our analysis reveals that current works, instead of building on previous findings in terms of temporal tasks and datasets, define their own tasks of temporal reasoning and create new datasets to solve them. Finally, we discuss how temporal reasoning evaluation can be advanced to better understand the temporal reasoning capabilities of language models.

Paper Type: Long

Research Area: Information Extraction

Research Area Keywords: Open information extraction

Contribution Types: Surveys

Languages Studied: English

Submission Number: 7545