Finding Sparse Structures for Domain Specific Neural Machine Translation

TL;DR

This paper introduces Prune-Tune, a novel method for domain-specific neural machine translation that learns sparse, disjoint sub-networks during fine-tuning, improving domain adaptation without degrading general performance.

Contribution

Prune-Tune is a new gradual pruning approach that creates multiple domain-specific sub-networks within a single model for effective multi-domain adaptation.

Findings

Outperforms strong baselines in target domain accuracy

Maintains general domain performance in single and multi-domain settings

Learns multiple disjoint sub-networks for different domains

Abstract

Neural machine translation often adopts the fine-tuning approach to adapt to specific domains. However, nonrestricted fine-tuning can easily degrade on the general domain and over-fit to the target domain. To mitigate the issue, we propose Prune-Tune, a novel domain adaptation method via gradual pruning. It learns tiny domain-specific sub-networks during fine-tuning on new domains. Prune-Tune alleviates the over-fitting and the degradation problem without model modification. Furthermore, Prune-Tune is able to sequentially learn a single network with multiple disjoint domain-specific sub-networks for multiple domains. Empirical experiment results show that Prune-Tune outperforms several strong competitors in the target domain test set without sacrificing the quality on the general domain in both single and multi-domain settings. The source code and data are available at https://github.com/ohlionel/Prune-Tune.