Learning to Plan and Generate Text with Citations

Anonymous

Abstract: The increasing demand for the deployment of LLMs in information-seeking scenarios has spurred efforts in creating verifiable systems, which generate responses to queries along with supporting evidence. In this paper, we explore the \emph{attribution} capabilities of plan-based models which have been recently shown to improve the faithfulness, grounding, and controllability of generated text. We conceptualize plans as a sequence of questions which serve as \emph{blueprints} of the generated content and its organization. We propose two attribution models that utilize different variants of blueprints, an \textit{abstractive} model where questions are generated from scratch, and an \textit{extractive} model where questions are copied from the input. Experiments on long-form question-answering show that planning consistently improves attribution quality. Moreover, the citations generated by blueprint models are more accurate compared to those obtained from LLM-based pipelines lacking a planning component.

Paper Type: long

Research Area: Summarization

Contribution Types: NLP engineering experiment

Languages Studied: English