Learning Evidence Highlighting for Frozen LLMs

TL;DR

HiLight is a framework that highlights key evidence in long contexts for frozen LLMs, improving reasoning without altering the original input.

Contribution

It introduces a reinforcement learning-based emphasis actor that highlights evidence, enhancing reasoning performance across tasks without modifying the LLM.

Findings

Consistently improves performance over prompt-based baselines.

Transfers zero-shot to unseen Solver models.

Does not require evidence labels or input modification.

Abstract

Large Language Models (LLMs) can reason well, yet often miss decisive evidence when it is buried in long, noisy contexts. We introduce HiLight, an Evidence Emphasis framework that decouples evidence selection from reasoning for frozen LLM solvers. HiLight avoids compressing or rewriting the input, which can discard or distort evidence, by training a lightweight Emphasis Actor to insert minimal highlight tags around pivotal spans in the unaltered context. A frozen Solver then performs downstream reasoning on the emphasized input. We cast highlighting as a weakly supervised decision-making problem and optimize the Actor with reinforcement learning using only the Solver's task reward, requiring no evidence labels and no access to or modification of the Solver. Across sequential recommendation and long-context question answering, HiLight consistently improves performance over strong prompt-based and automated prompt-optimization baselines. The learned emphasis policy transfers zero-shot to both smaller and larger unseen Solver families, including an API-based Solver, suggesting that the Actor captures genuine, reusable evidence structure rather than overfitting to a single backbone.