A Taxonomy for Data Contamination in Large Language Models
This addresses a growing concern for researchers and practitioners in NLP by providing a framework to understand and mitigate contamination effects, though it is incremental as it builds on existing decontamination efforts.
The paper tackles the problem of data contamination in large language models, where evaluation datasets may be present in pretraining data, inflating performance; it presents a taxonomy to categorize contamination types and analyzes their impact on summarization and question answering tasks, revealing which types pose the highest risk.
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.