Retrieval-augmented episodic memory narrows LLMs' lexical frequency gap in syntax tests
najoungkim · x · 2026-10-02
A paper by Najoung Kim and Jing Liu (Boston University) tests whether episodic memory can fix LLMs' sensitivity to lexical frequency in grammaticality judgments. Using k-NN language models—explicit instance storage bolted onto parametric models—as an instantiation of hippocampal episodic memory, they show retrieval augmentation narrows the performance gap between high- and low-frequency items in frequency-stratified syntactic contrast tests, consistently across structures. The work will be presented at COLM 2026.
More from Research
- CAS Five-Year Plan gives AI for Science its own chapter, setting up a US-vs-China metascience bet — teortaxesTex · 2026-10-03
- Draft standard for autonomous labs emerges as AI-driven science heats up — Afinetheorem · 2026-10-03
- Papers with 'agent' in the title get cited more: AI lit search reads titles too — rajammanabrolu · 2026-10-03
- EgoTools: 100-hour egocentric video dataset teaches AI tool-centric reasoning — liuziwei7 · 2026-10-03
- Nat Lambert co-founds Trillium Labs, a nonprofit for open frontier AI science — natolambert · 2026-10-03
- SWE-chat V2 ships 3.5x larger with agent skills and subagent trajectories — Diyi_Yang · 2026-10-03