A Taxonomy for Data Contamination in Large Language Models

TL;DR

This paper introduces a taxonomy categorizing data contamination types in large language models, analyzing their impact on downstream tasks like summarization and question answering to understand contamination effects.

Contribution

It provides a novel taxonomy for data contamination in LLMs and analyzes how different contamination types affect model performance on key NLP tasks.

Findings

Certain contamination types significantly inflate model performance.

Contamination from test set variants can evade detection.

Impact varies across different NLP tasks.

Abstract

Large language models pretrained on extensive web corpora demonstrate remarkable performance across a wide range of downstream tasks. However, a growing concern is data contamination, where evaluation datasets may be contained in the pretraining corpus, inflating model performance. Decontamination, the process of detecting and removing such data, is a potential solution; yet these contaminants may originate from altered versions of the test set, evading detection during decontamination. How different types of contamination impact the performance of language models on downstream tasks is not fully understood. We present a taxonomy that categorizes the various types of contamination encountered by LLMs during the pretraining phase and identify which types pose the highest risk. We analyze the impact of contamination on two key NLP tasks -- summarization and question answering -- revealing how different types of contamination influence task performance during evaluation.