Donkii: Can Annotation Error Detection Methods Find Errors in Instruction-Tuning Datasets?

TL;DR

This paper introduces DONKII, a novel benchmark for annotation error detection in instruction-tuning datasets for language models, demonstrating the importance of AED methods in improving data quality and model performance.

Contribution

It presents the first benchmark dataset for AED in instruction tuning, along with a taxonomy of error types and evaluation of AED methods for language generation tasks.

Findings

All datasets contain clear errors that can propagate into models

Model size and AED method choice significantly affect error detection performance

Practical recommendations for using AED to improve instruction-tuning data

Abstract

Instruction tuning has become an integral part of training pipelines for Large Language Models (LLMs) and has been shown to yield strong performance gains. In an orthogonal line of research, Annotation Error Detection (AED) has emerged as a tool for detecting quality problems in gold standard labels. So far, however, the application of AED methods has been limited to classification tasks. It is an open question how well AED methods generalize to language generation settings, which are becoming more widespread via LLMs. In this paper, we present a first and novel benchmark for AED on instruction tuning data: DONKII. It comprises three instruction-tuning datasets enriched with error annotations by experts and semi-automatic methods. We also provide a novel taxonomy of error types for instruction-tuning data. We find that all three datasets contain clear errors, which sometimes propagate directly into instruction-tuned LLMs. We propose four AED baselines for the generative setting and evaluate them extensively on the newly introduced dataset. Our results show that the choice of the right AED method and model size is indeed crucial and derive practical recommendations for how to use AED methods to clean instruction-tuning data.