To Copy or Not to Copy: Copying Is Easier to Induce Than Recall

arXiv:2601.12075v2 Announce Type: replace Abstract: Language models used in retrieval-augmented settings must arbitrate between parametric knowledge stored in their weights and contextual information in the prompt. This work presents a mechanistic study of that choice by extracting an arbitration vector from model activations on a curated dataset designed to disentangle (i) irrelevant contexts that elicit parametric recall and (ii) relevant but false contexts that elicit copying. The vector is computed as the residual-stream centroid difference between these regimes across 27 relations, and is injected as an additive intervention at selected layers and token spans to steer behavior in two directions: CopyrightarrowRecall (suppressing context use) and RecallrightarrowCopy (inducing the model to copy any token from the context). Experiments on three architectures (decoder-only and encoder/decoder) and two open-domain QA benchmarks show consistent behavior shifts under moderate scaling while monitoring accuracy and fluency. Mechanistic analyses of attention routing, MLP contributions, and layer-wise probability trajectories reveal an asymmetry: inducing copying is an easy reactivation'' process that can be triggered at different locations in the input, while restoring recall is a suppression'' process that is more fragile and strongly tied to object-token interventions.

Related

• Retrieval Heads are Dynamic
• In-depth Analysis of Graph-based RAG in a Unified Framework
• Through a Compressed Lens: Investigating The Impact of Quantization on Factual Knowledge Recall

Source: arXiv cs.CL | 2026-09-01