Temporal Understanding under Deictic Frame of Reference
Damin Zhang, Julia Rayz
TL;DR
The paper introduces TUuD, a framework to evaluate how LLMs reason about time from a deictic, experiencer-based perspective by shifting the reference point of $t_{now}$ and measuring similarity between $t_{now}$ and target events across relative distance $dist$. Using a dataset derived from Wikipedia timelines and TimeQA annotations, TUuD evaluates four LLMs and reveals a clear human-like adaptation pattern: similarity peaks near the present and declines with increasing temporal distance, though robustness diminishes beyond about 50 years. The work demonstrates that LLMs can partially internalize deictic temporal relations but remain sensitive to reference-frame shifts and temporal distance, with model-specific deviations and effects from anonymization, rating ranges, and timestamp perturbations. These findings suggest that stronger temporal cognition in LLMs will require grounded grounding, temporally aware training, and explicit frame-of-reference modeling to achieve robust performance across diverse temporal contexts.
Abstract
Understanding time is fundamental to human cognition, where temporal experience is often conceptualized through spatial metaphors grounded in sensory-motor experience. For example, "summer is approaching" parallels "We are approaching the summer". In such expressions, humans rely on a frame of reference (FoR) to interpret meaning relative to a particular viewpoint. Extending this concept to time, a temporal frame of reference (t-FoR) defines how temporal relations are perceived relative to an experiencer's moment of "now". While Large Language Models (LLMs) have shown remarkable advances in natural language understanding, their ability to interpret and reason about time remains limited. In this work, we introduce TUuD (Temporal Understanding under Deictic t-FoR), a framework that evaluates how LLMs interpret time-event and event-event relations when the reference point of "now" dynamically shifts along a timeline. Following recent work on temporal cognition \cite{li2025other}, LLMs are prompted to rate the similarity between the current moment and a target event from 0.00 (completely dissimilar) to 1.00 (highly similar), where similarity quantifies perceived temporal alignment between the two points. Our results show that four evaluated LLMs exhibit measurable adaptation to a deictic t-FoR, with similarity ratings peaking around the present and decreasing toward past and future events. The adaptation, however, weakens beyond near-term contexts, suggesting that while LLMs display partial human-like temporal cognition, their temporal reasoning remains sensitive to reference-frame shifts and temporal distance.