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

arXiv:2601.12075 · cs.CL · Submitted 2026-01-17 · Read on arXiv

cs.CL

Submitted: 2026-01-17

Updated: 2026-08-31

Comments: Accepted at EMNLP 2026

Code: https://github.com/m3hrdadfi/copy-or-not-to-copy

License: http://creativecommons.org/licenses/by-sa/4.0/

The gist: Language models used in retrieval-augmented settings must arbitrate between parametric knowledge stored in their weights and contextual information in the prompt.

Terminology

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: Copy to Recall (suppressing context use) and Recall to Copy (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.

Sources

Related papers