A Taxonomy for Data Contamination in Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Palavalli, Medha, Bertsch, Amanda, Gormley, Matthew R.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910523277180928
author Palavalli, Medha
Bertsch, Amanda
Gormley, Matthew R.
author_facet Palavalli, Medha
Bertsch, Amanda
Gormley, Matthew R.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2407_08716
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle A Taxonomy for Data Contamination in Large Language Models
Palavalli, Medha
Bertsch, Amanda
Gormley, Matthew R.
Computation and Language
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.
title A Taxonomy for Data Contamination in Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2407.08716