Mitigating Provenance-Role Collapse in Long-Term Agents via Typed Memory Representation
2026-05-25 • Computation and Language
Computation and Language
AI summaryⓘ
The authors explain that current large language model (LLM) agents keep their memories as plain text, which can cause confusion about where information comes from. To fix this, they created MemIR, a new way to organize memory by separating facts, evidence, and cues to clearly track sources. This method helps the agent check facts better and handle complicated questions involving time and multiple pieces of evidence. Tests showed that MemIR works better than older methods, especially when it’s important to know the origin of information.
long-term memorylarge language modelssource monitoringMemory Intermediate Representationprovenanceatomic projectionretrieval cuesfactual authorizationtemporal groundingevidence aggregation
Authors
Zhengda Jin, Bingbing Wang, Jing Li, Ruifeng Xu, Min Zhang
Abstract
Long-term memory is essential for persistent LLM agents, yet prevailing architectures store historical interactions as unstructured, flat text. This unconstrained storage induces provenance-role collapse, a critical failure mode where agents suffer from source-monitoring errors. To resolve this cognitive vulnerability at the architectural level, we propose MemIR, a typed Memory Intermediate Representation that operationalizes source monitoring as a structural constraint. MemIR writes long-term memory into grounded atoms that separate raw evidence, retrieval cues, and truth-bearing claims, with factual authorization restricted to supported claim atoms. It then applies multi-route atomic projection and provenance-scoped utilization to transform heterogeneous retrieval hits into claim-centered candidate bundles and a normalized fact interface for answer generation. Experiments on LoCoMo and BEAM-100K demonstrate that MemIR consistently outperforms existing memory baselines, especially on tasks requiring source tracking, temporal grounding, and aggregation of fragmented evidence.