From Ground Truth to Measurement: A Statistical Framework for Human Labeling
This work addresses the issue of systematic variation in human labeling for machine learning practitioners, offering a diagnostic tool to improve data quality and model understanding, though it is incremental in extending existing measurement-error models.
The paper tackles the problem of human labeling variation in supervised machine learning by reframing annotation as a measurement process, introducing a statistical framework to decompose labeling outcomes into interpretable sources like instance difficulty and annotator bias, and applying it to a natural language inference dataset to find empirical evidence for these components.
Supervised machine learning assumes that labeled data provide accurate measurements of the concepts models are meant to learn. Yet in practice, human labeling introduces systematic variation arising from ambiguous items, divergent interpretations, and simple mistakes. Machine learning research commonly treats all disagreement as noise, which obscures these distinctions and limits our understanding of what models actually learn. This paper reframes annotation as a measurement process and introduces a statistical framework for decomposing labeling outcomes into interpretable sources of variation: instance difficulty, annotator bias, situational noise, and relational alignment. The framework extends classical measurement-error models to accommodate both shared and individualized notions of truth, reflecting traditional and human label variation interpretations of error, and provides a diagnostic for assessing which regime better characterizes a given task. Applying the proposed model to a multi-annotator natural language inference dataset, we find empirical evidence for all four theorized components and demonstrate the effectiveness of our approach. We conclude with implications for data-centric machine learning and outline how this approach can guide the development of a more systematic science of labeling.