Factuality on Demand: Controlling the Factuality-Informativeness Trade-off in Text Generation

TL;DR

This paper introduces Factuality-Controlled Generation (FCG), a framework allowing users to specify factuality constraints in text generation, balancing factual accuracy and informativeness in responses.

Contribution

The paper proposes a novel FCG framework, synthetic training data for models, and evaluation methods for factuality and informativeness trade-offs.

Findings

Synthetic training improves factuality adherence

Models better balance factuality and informativeness

Framework adaptable to different application needs

Abstract

Large language models (LLMs) encode knowledge with varying degrees of confidence. When responding to queries, models face an inherent trade-off: they can generate responses that are less informative but highly factual, or more informative but potentially less accurate. Different applications demand different balances between informativeness and factuality. We introduce Factuality-Controlled Generation (FCG), a framework that enables users to specify factuality constraints alongside their queries. We propose to evaluate FCG performance on two dimensions: adherence to factuality constraints and response informativeness. We propose to train models on the FCG task using synthetic data, and show that our synthetic training significantly improves models' ability to both respect factuality requirements and maintain informativeness in their outputs.