CoCoLex: Confidence-guided Copy-based Decoding for Grounded Legal Text Generation

TL;DR

CoCoLex is a decoding strategy for legal text generation that enhances faithfulness by dynamically copying from context based on model confidence, outperforming existing methods on multiple benchmarks.

Contribution

Introduces CoCoLex, a novel confidence-guided copy decoding method that improves fidelity in legal text generation by effectively integrating external context.

Findings

Outperforms existing context-aware decoding methods

Enhances faithfulness in long-form legal text generation

Demonstrates effectiveness on five legal benchmarks

Abstract

Due to their ability to process long and complex contexts, LLMs can offer key benefits to the Legal domain, but their adoption has been hindered by their tendency to generate unfaithful, ungrounded, or hallucinatory outputs. While Retrieval-Augmented Generation offers a promising solution by grounding generations in external knowledge, it offers no guarantee that the provided context will be effectively integrated. To address this, context-aware decoding strategies have been proposed to amplify the influence of relevant context, but they usually do not explicitly enforce faithfulness to the context. In this work, we introduce Confidence-guided Copy-based Decoding for Legal Text Generation (CoCoLex)-a decoding strategy that dynamically interpolates the model produced vocabulary distribution with a distribution derived based on copying from the context. CoCoLex encourages direct copying based on the model's confidence, ensuring greater fidelity to the source. Experimental results on five legal benchmarks demonstrate that CoCoLex outperforms existing context-aware decoding methods, particularly in long-form generation tasks.