Verifiable Generation with Subsentence-Level Fine-Grained Citations
Shuyang Cao, Lu Wang
TL;DR
This work tackles verifiable generation by enforcing subsentence-level, fine-grained citations and introducing SCiFi, a 10K-paragraph Wikipedia-based dataset with candidate sources and query guidance. It benchmarks multiple LLMs and document-reading strategies, demonstrating that access to complete source context improves citation quality, while larger models enhance answer quality but do not consistently improve citations. Supervised fine-tuning emerges as necessary to increase subsentence-level citation density, underscoring the need for task-specific training to align outputs with supporting documents. The dataset and findings advance transparent AI by clarifying how context and training shape precise attribution, with practical implications for trustworthy information synthesis and documentation.
Abstract
Verifiable generation requires large language models (LLMs) to cite source documents supporting their outputs, thereby improve output transparency and trustworthiness. Yet, previous work mainly targets the generation of sentence-level citations, lacking specificity about which parts of a sentence are backed by the cited sources. This work studies verifiable generation with subsentence-level fine-grained citations for more precise location of generated content supported by the cited sources. We first present a dataset, SCiFi, comprising 10K Wikipedia paragraphs with subsentence-level citations. Each paragraph is paired with a set of candidate source documents for citation and a query that triggers the generation of the paragraph content. On SCiFi, we evaluate the performance of state-of-the-art LLMs and strategies for processing long documents designed for these models. Our experiment results reveals key factors that could enhance the quality of citations, including the expansion of the source documents' context accessible to the models and the implementation of specialized model tuning.