NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark


In this position paper, we argue that the classical evaluation on Natural Language Processing (NLP) tasks using annotated benchmarks is in trouble. The worst kind of data contamination happens when a Large Language Model (LLM) is trained on the test split of a benchmark, and then evaluated in the same benchmark. The extent of the problem is unknown, as it is not straightforward to measure. Contamination causes an overestimation of the performance of a contaminated model in a target benchmark and associated task with respect to their non-contaminated counterparts. The consequences can be very harmful, with wrong scientific conclusions being published while other correct ones are discarded. This position paper defines different levels of data contamination and argues for a community effort, including the development of automatic and semi-automatic measures to detect when data from a benchmark was exposed to a model, and suggestions for flagging papers with conclusions that are compromised by data contamination.

EMNLP 2023 Findings
Julen Etxaniz
Julen Etxaniz
Estudiante de Doctorado en Análisis y Procesamiento del Lenguaje

Estudiante de Doctorado en Análisis y Procesamiento del Lenguaje en HiTZ Center IXA Group (UPV/EHU). Trabajando en mejorar los modelos de lenguaje para idiomas con pocos recursos. Graduado en Ingeniería Informática con especialidad en Ingeniería del Software. Máster en Análisis y Procesamiento del Lenguaje.