Papers › Linear Correlation in LM's Compositional Generalization and Hallucination

Linear Correlation in LM's Compositional Generalization and Hallucination

6 Feb 2025arXiv:2502.04520archive 2025-07-28

Letian Peng, Chenyang An, Shibo Hao, chengyu dong, Jingbo Shang

The generalization of language models (LMs) is undergoing active debates, contrasting their potential for general intelligence with their struggles with basic knowledge composition (e.g., reverse/transition curse). This paper uncovers the phenomenon of linear correlations in LMs during knowledge composition. For explanation, there exists a linear transformation between certain related knowledge that maps the next token prediction logits from one prompt to another, e.g., "X lives in the city of" → "X lives in the country of" for every given X. This mirrors the linearity in human knowledge composition, such as Paris → France. Our findings indicate that the linear transformation is resilient to large-scale fine-tuning, generalizing updated knowledge when aligned with real-world relationships, but causing hallucinations when it deviates. Empirical results suggest that linear correlation can serve as a potential identifier of LM's generalization. Finally, we show such linear correlations can be learned with a single feedforward network and pre-trained vocabulary representations, indicating LM generalization heavily relies on the latter.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

KomeijiForce/LinCorr officialmentioned on GitHub report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Hallucination

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Dense ConnectionsFeedforward Network

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections