SelfCite: Self-Supervised Alignment for Context Attribution in Large Language Models
This addresses the issue of improving citation accuracy in LLMs for long-form question answering, which is incremental as it builds on existing methods with a novel self-supervised technique.
The paper tackles the problem of generating high-quality, sentence-level citations for statements in LLM responses by introducing SelfCite, a self-supervised approach that uses context ablation to provide a reward signal, resulting in an increase in citation F1 by up to 5.3 points on the LongBench-Cite benchmark.
We introduce SelfCite, a novel self-supervised approach that aligns LLMs to generate high-quality, fine-grained, sentence-level citations for the statements in their generated responses. Instead of only relying on costly and labor-intensive annotations, SelfCite leverages a reward signal provided by the LLM itself through context ablation: If a citation is necessary, removing the cited text from the context should prevent the same response; if sufficient, retaining the cited text alone should preserve the same response. This reward can guide the inference-time best-of-N sampling strategy to improve citation quality significantly, as well as be used in preference optimization to directly fine-tune the models for generating better citations. The effectiveness of SelfCite is demonstrated by increasing citation F1 up to 5.3 points on the LongBench-Cite benchmark across five long-form question answering tasks. The source code is available at https://github.com/facebookresearch/SelfCite