Data Checklist: On Unit-Testing Datasets with Usable Information

TL;DR

This paper introduces a principled, taxonomy-based approach to unit-testing datasets for language models, enabling detection of known and unknown artifacts and improving data efficiency in model alignment.

Contribution

It proposes a novel taxonomy for dataset unit-testing, called data checklists, which systematically identify artifacts and enhance data filtering for better model alignment.

Findings

Recovered known artifacts in SNLI dataset

Discovered new artifacts in LLM preference datasets

Improved data filtering enhances alignment efficacy

Abstract

Model checklists (Ribeiro et al., 2020) have emerged as a useful tool for understanding the behavior of LLMs, analogous to unit-testing in software engineering. However, despite datasets being a key determinant of model behavior, evaluating datasets, e.g., for the existence of annotation artifacts, is largely done ad hoc, once a problem in model behavior has already been found downstream. In this work, we take a more principled approach to unit-testing datasets by proposing a taxonomy based on the V-information literature. We call a collection of such unit tests a data checklist. Using a checklist, not only are we able to recover known artifacts in well-known datasets such as SNLI, but we also discover previously unknown artifacts in preference datasets for LLM alignment. Data checklists further enable a new kind of data filtering, which we use to improve the efficacy and data efficiency of preference alignment.