🤖 AI Summary
This work addresses inter-annotator scale inconsistency in subjective annotation tasks. We propose a post-hoc recalibration method leveraging natural language explanations: the original Likert-scale ratings and their associated free-text justifications are jointly fed into a large language model (LLM), which dynamically remaps them to a unified numerical scale—without requiring predefined error categories or fixed rating guidelines. Our key contribution is the first integration of explanatory text with ordinal ratings for post-hoc recalibration, enabling flexible, annotation-time-agnostic revision of rating criteria. Empirical evaluation on document-level question answering assessment demonstrates that the recalibrated scores better align with expert judgments, preserve inter-annotator agreement, and achieve near-human performance in LLM-generated scoring.
📝 Abstract
The rise of large language models (LLMs) has brought a critical need for high-quality human-labeled data, particularly for processes like human feedback and evaluation. A common practice is to label data via consensus annotation over human judgments. However, annotators' judgments for subjective tasks can differ in many ways: they may reflect different qualitative judgments about an example, and they may be mapped to a labeling scheme in different ways. We show that these nuances can be captured by natural language explanations, and propose a method to rescale ordinal annotations and explanations using LLMs. Specifically, we feed annotators' Likert ratings and corresponding explanations into an LLM and prompt it to produce a numeric score anchored in a scoring rubric. These scores should reflect the annotators' underlying assessments of the example. The rubric can be designed or modified after annotation, and include distinctions that may not have been known when the original error taxonomy was devised. We explore our technique in the context of rating system outputs for a document-grounded question answering task, where LLMs achieve near-human performance. Our method rescales the raw judgments without impacting agreement and brings the scores closer to human judgments grounded in the same scoring rubric.
