Investigating Catastrophic Forgetting During Continual Training for Neural Machine Translation

TL;DR

This paper investigates the causes of catastrophic forgetting in neural machine translation during continual training, focusing on modules and parameters, and provides insights into how different parts of the model contribute to knowledge retention and loss.

Contribution

It offers a detailed analysis of module and parameter roles in catastrophic forgetting, advancing understanding of NMT model behavior during domain adaptation.

Findings

Some modules are linked to general knowledge, others to domain adaptation.

Certain parameters are crucial for both general and in-domain translation.

Significant changes in key parameters cause performance decline.

Abstract

Neural machine translation (NMT) models usually suffer from catastrophic forgetting during continual training where the models tend to gradually forget previously learned knowledge and swing to fit the newly added data which may have a different distribution, e.g. a different domain. Although many methods have been proposed to solve this problem, we cannot get to know what causes this phenomenon yet. Under the background of domain adaptation, we investigate the cause of catastrophic forgetting from the perspectives of modules and parameters (neurons). The investigation on the modules of the NMT model shows that some modules have tight relation with the general-domain knowledge while some other modules are more essential in the domain adaptation. And the investigation on the parameters shows that some parameters are important for both the general-domain and in-domain translation and the great change of them during continual training brings about the performance decline in general-domain. We conduct experiments across different language pairs and domains to ensure the validity and reliability of our findings.