Extent of Explicit Temporal Representation Needed for Cross-Session Reasoning

Determine the degree of explicit temporal representation required within long-term conversational memory architectures for large language model–based agents to achieve accurate cross-session reasoning over multi-session dialogue histories.

Background

The paper reviews two primary benchmarks for long-term conversational memory—LongMemEvalS and LoCoMo—and notes that while they evaluate various aspects of memory, they do not isolate the effect of temporal structuring. As a result, it is not established how much explicit temporal representation contributes to accurate reasoning across sessions.

This gap motivates Chronos, which selectively structures temporally grounded events while preserving raw dialogue. However, the authors explicitly state that it remains unresolved how much explicit temporal representation is actually necessary for accurate cross-session reasoning, independent of other system components.