Relay Decoding: Concatenating Large Language Models for Machine Translation

TL;DR

This paper introduces Relay Decoding, a novel method that concatenates two large language models supporting different languages for machine translation, reducing costs and improving performance with limited parallel data.

Contribution

The paper proposes RD, a new approach that connects separate language models via a mapping layer for efficient multilingual translation.

Findings

Achieves superior translation results on Multi30k and WikiMatrix datasets.

Reduces the need for large multilingual models by concatenating monolingual models.

Demonstrates effectiveness with limited parallel data.

Abstract

Leveraging large language models for machine translation has demonstrated promising results. However, it does require the large language models to possess the capability of handling both the source and target languages in machine translation. When it is challenging to find large models that support the desired languages, resorting to continuous learning methods becomes a costly endeavor. To mitigate these expenses, we propose an innovative approach called RD (Relay Decoding), which entails concatenating two distinct large models that individually support the source and target languages. By incorporating a simple mapping layer to facilitate the connection between these two models and utilizing a limited amount of parallel data for training, we successfully achieve superior results in the machine translation task. Experimental results conducted on the Multi30k and WikiMatrix datasets validate the effectiveness of our proposed method.