More Parameters? No Thanks!

TL;DR

This paper introduces a parameter-free adaptation method for multilingual neural machine translation that prunes and retrains redundant parameters to improve bilingual performance and reduce negative interference without additional modules.

Contribution

It proposes a novel iterative pruning and retraining strategy that leverages redundant parameters to enhance multilingual translation quality.

Findings

Pruning 50-70% of parameters causes minimal BLEU score drop.

The method improves high-resource language translation by an average of +1.36 BLEU.

Redundancies exist in MNMT models, enabling efficient adaptation.

Abstract

This work studies the long-standing problems of model capacity and negative interference in multilingual neural machine translation MNMT. We use network pruning techniques and observe that pruning 50-70% of the parameters from a trained MNMT model results only in a 0.29-1.98 drop in the BLEU score. Suggesting that there exist large redundancies even in MNMT models. These observations motivate us to use the redundant parameters and counter the interference problem efficiently. We propose a novel adaptation strategy, where we iteratively prune and retrain the redundant parameters of an MNMT to improve bilingual representations while retaining the multilinguality. Negative interference severely affects high resource languages, and our method alleviates it without any additional adapter modules. Hence, we call it parameter-free adaptation strategy, paving way for the efficient adaptation of MNMT. We demonstrate the effectiveness of our method on a 9 language MNMT trained on TED talks, and report an average improvement of +1.36 BLEU on high resource pairs. Code will be released here.