Memorization vs. Generalization: Quantifying Data Leakage in NLP Performance Evaluation
This work addresses the critical problem of accurately evaluating NLP model generalization for researchers and practitioners by exposing flaws in common benchmark datasets.
This paper investigates data leakage in public NLP datasets, finding that overlap between train and test sets can inflate performance metrics. The study assesses how this leakage impacts a model's ability to memorize versus generalize on tasks like named entity recognition and relation extraction.
Public datasets are often used to evaluate the efficacy and generalizability of state-of-the-art methods for many tasks in natural language processing (NLP). However, the presence of overlap between the train and test datasets can lead to inflated results, inadvertently evaluating the model's ability to memorize and interpreting it as the ability to generalize. In addition, such data sets may not provide an effective indicator of the performance of these methods in real world scenarios. We identify leakage of training data into test data on several publicly available datasets used to evaluate NLP tasks, including named entity recognition and relation extraction, and study them to assess the impact of that leakage on the model's ability to memorize versus generalize.