Temporal Context Reinstatement Drives Episodic-Like Order Memory in Long-Context Language Models

This work identifies internal computations that support episodic-like temporal order judgments in long-context language models. We find a compact temporal code and a time-reinstatement attention head whose causal manipulation changes memory performance.

Recommended citation: Pink, M., Vo, V. A., Wu, Q., Mu, J., Turek, J. S., Hasson, U., Norman, K. A., Michelmann, S., Huth, A., & Toneva, M. (2026). "Temporal Context Reinstatement Drives Episodic-Like Order Memory in Long-Context Language Models." Forty-third International Conference on Machine Learning (ICML 2026). https://arxiv.org/abs/2607.22575