CABLE: Extending the Reach of Memory Retrieval via Complementary Antecedent-Based Linking and Expansion
CABLE adds sparse memory links meant to recover evidence that semantic retrieval misses.
The paper frames the problem as evidence reachability in long-term conversational memory, where relevant earlier plans, motivations, or experiences may be semantically distant from later queries. CABLE generates antecedent-oriented queries for each new memory, filters out candidates already close to the retriever’s semantic neighborhood, verifies what remains, and stores accepted links in a directed graph. At retrieval time, it expands from the host retriever’s seeds along those links to surface supporting evidence. The authors report higher mean LLM-judge scores across tested memory systems, with the largest gains on open-domain, multi-session, and preference-oriented questions. ArXiv · AI/CL/LG's note
The paper frames the problem as evidence reachability in long-term conversational memory, where relevant earlier plans, motivations, or experiences may be semantically distant from later queries. CABLE generates antecedent-oriented queries for each new memory, filters out candidates already close to the retriever’s semantic neighborhood, verifies what remains, and stores accepted links in a directed graph. At retrieval time, it expands from the host retriever’s seeds along those links to surface supporting evidence. The authors report higher mean LLM-judge scores across tested memory systems, with the largest gains on open-domain, multi-session, and preference-oriented questions. ArXiv · AI/CL/LG's note
score 5