DnDScore: Decontextualization and Decomposition for Factuality Verification in Long-Form Text Generation

TL;DR

This paper introduces DnDScore, a new verification method that combines decontextualization and decomposition strategies to improve factuality verification in long-form text generated by large language models.

Contribution

It systematically investigates the interaction between decomposition and decontextualization, and proposes DnDScore, a novel verification approach that accounts for decontextualization effects.

Findings

Strategy choice significantly impacts factuality scores.

Decontextualization-aware verification improves accuracy.

Evaluation of different methods highlights their interactions.

Abstract

The decompose-then-verify strategy for verification of Large Language Model (LLM) generations decomposes claims that are then independently verified. Decontextualization augments text (claims) to ensure it can be verified outside of the original context, enabling reliable verification. While decomposition and decontextualization have been explored independently, their interactions in a complete system have not been investigated. Their conflicting purposes can create tensions: decomposition isolates atomic facts while decontextualization inserts relevant information. Furthermore, a decontextualized subclaim presents a challenge to the verification step: what part of the augmented text should be verified as it now contains multiple atomic facts? We conduct an evaluation of different decomposition, decontextualization, and verification strategies and find that the choice of strategy matters in the resulting factuality scores. Additionally, we introduce DnDScore, a decontextualization aware verification method which validates subclaims in the context of contextual information.