Overestimation in LLM Evaluation: A Controlled Large-Scale Study on Data Contamination’s Impact on Machine Translation

Muhammed Yusuf Kocyigit, Eleftheria Briakou, Daniel Deutsch, Jiaming Luo, Colin Cherry, Markus Freitag
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:31105-31132, 2025.

Abstract

Data contamination—the accidental consumption of evaluation examples within the pre-training data—can undermine the validity of evaluation benchmarks. In this paper, we present a rigorous analysis of the effects of contamination on language models at 1B and 8B scales on the machine translation task. Starting from a carefully decontaminated train-test split, we systematically introduce contamination at various stages, scales, and data formats to isolate its effect and measure its impact on performance metrics. Our experiments reveal that contamination with both source and target substantially inflates BLEU scores, and this inflation is 2.5 times larger (up to 30 BLEU points) for 8B compared to 1B models. In contrast, source-only and target-only contamination generally produce smaller, less consistent over-estimations. Finally, we study how the temporal distribution and frequency of contaminated samples influence performance over-estimation across languages with varying degrees of data resources.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-kocyigit25a, title = {Overestimation in {LLM} Evaluation: A Controlled Large-Scale Study on Data Contamination’s Impact on Machine Translation}, author = {Kocyigit, Muhammed Yusuf and Briakou, Eleftheria and Deutsch, Daniel and Luo, Jiaming and Cherry, Colin and Freitag, Markus}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {31105--31132}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/kocyigit25a/kocyigit25a.pdf}, url = {https://proceedings.mlr.press/v267/kocyigit25a.html}, abstract = {Data contamination—the accidental consumption of evaluation examples within the pre-training data—can undermine the validity of evaluation benchmarks. In this paper, we present a rigorous analysis of the effects of contamination on language models at 1B and 8B scales on the machine translation task. Starting from a carefully decontaminated train-test split, we systematically introduce contamination at various stages, scales, and data formats to isolate its effect and measure its impact on performance metrics. Our experiments reveal that contamination with both source and target substantially inflates BLEU scores, and this inflation is 2.5 times larger (up to 30 BLEU points) for 8B compared to 1B models. In contrast, source-only and target-only contamination generally produce smaller, less consistent over-estimations. Finally, we study how the temporal distribution and frequency of contaminated samples influence performance over-estimation across languages with varying degrees of data resources.} }
Endnote
%0 Conference Paper %T Overestimation in LLM Evaluation: A Controlled Large-Scale Study on Data Contamination’s Impact on Machine Translation %A Muhammed Yusuf Kocyigit %A Eleftheria Briakou %A Daniel Deutsch %A Jiaming Luo %A Colin Cherry %A Markus Freitag %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-kocyigit25a %I PMLR %P 31105--31132 %U https://proceedings.mlr.press/v267/kocyigit25a.html %V 267 %X Data contamination—the accidental consumption of evaluation examples within the pre-training data—can undermine the validity of evaluation benchmarks. In this paper, we present a rigorous analysis of the effects of contamination on language models at 1B and 8B scales on the machine translation task. Starting from a carefully decontaminated train-test split, we systematically introduce contamination at various stages, scales, and data formats to isolate its effect and measure its impact on performance metrics. Our experiments reveal that contamination with both source and target substantially inflates BLEU scores, and this inflation is 2.5 times larger (up to 30 BLEU points) for 8B compared to 1B models. In contrast, source-only and target-only contamination generally produce smaller, less consistent over-estimations. Finally, we study how the temporal distribution and frequency of contaminated samples influence performance over-estimation across languages with varying degrees of data resources.
APA
Kocyigit, M.Y., Briakou, E., Deutsch, D., Luo, J., Cherry, C. & Freitag, M.. (2025). Overestimation in LLM Evaluation: A Controlled Large-Scale Study on Data Contamination’s Impact on Machine Translation. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:31105-31132 Available from https://proceedings.mlr.press/v267/kocyigit25a.html.

Related Material