ContextCite: Attributing Model Generation to Context

Benjamin Cohen-Wang, Harshay Shah, Kristian Georgiev, Aleksander Madry

Keywords: attribution, citation, generative models, large language models

TL;DR: We introduce ContextCite, a simple and scalable method to identify the parts of the context that are responsible for a language model generating a particular statement.

Abstract:

How do language models use information provided as context when generating a response?
Can we infer whether a particular generated statement is actually grounded in the context, a misinterpretation, or fabricated?
To help answer these questions, we introduce the problem of *context attribution*: pinpointing the parts of the context (if any) that *led* a model to generate a particular statement.
We then present ContextCite, a simple and scalable method for context attribution that can be applied on top of any existing language model.
Finally, we showcase the utility of ContextCite through three applications:
(1) helping verify generated statements
(2) improving response quality by pruning the context and
(3) detecting poisoning attacks.
We provide code for ContextCite at https://github.com/MadryLab/context-cite.

Primary Area: Natural language processing

Submission Number: 17808