Attributed Question Answering: Evaluation and Modeling for Attributed Large Language Models

TL;DR

This paper introduces Attributed Question Answering, proposing an evaluation framework and benchmarking architectures to assess how well large language models can attribute generated text, which is crucial for information-seeking applications.

Contribution

It formulates Attributed QA as a new task, develops a reproducible evaluation framework, and benchmarks various architectures for attribution capabilities.

Findings

A correlated automatic metric effectively measures attribution.

Current state-of-the-art methods show varying performance on attribution tasks.

Insights into building LLMs with better attribution capabilities.

Abstract

Large language models (LLMs) have shown impressive results while requiring little or no direct supervision. Further, there is mounting evidence that LLMs may have potential in information-seeking scenarios. We believe the ability of an LLM to attribute the text that it generates is likely to be crucial in this setting. We formulate and study Attributed QA as a key first step in the development of attributed LLMs. We propose a reproducible evaluation framework for the task and benchmark a broad set of architectures. We take human annotations as a gold standard and show that a correlated automatic metric is suitable for development. Our experimental work gives concrete answers to two key questions (How to measure attribution?, and How well do current state-of-the-art methods perform on attribution?), and give some hints as to how to address a third (How to build LLMs with attribution?).