An empirical study of memorization in NLP

• Main Authors: ZHENG, Xiaosen, JIANG, Jing
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2022
• Subjects: Natural language processing; Computer Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/7705; https://ink.library.smu.edu.sg/context/sis_research/article/8708/viewcontent/2022.acl_long.434.pdf
• Institution: Singapore Management University

A recent study by Feldman (2020) proposed a long-tail theory to explain the memorization behavior of deep learning models. However, memorization has not been empirically verified in the context of NLP, a gap addressed by this work. In this paper, we use three different NLP tasks to check if the long-tail theory holds. Our experiments demonstrate that top-ranked memorized training instances are likely atypical, and removing the top-memorized training instances leads to a more serious drop in test accuracy compared with removing training instances randomly. Furthermore, we develop an attribution method to better understand why a training instance is memorized. We empirically show that our memorization attribution method is faithful, and share our interesting finding that the top-memorized parts of a training instance tend to be features negatively correlated with the class label.