Core Concepts
EBBS is a novel ensemble method with a bi-level beam search algorithm that outperforms direct and pivot translations, improving translation quality in zero-shot machine translation.
Abstract
EBBS introduces a unique approach to zero-shot machine translation by utilizing an ensemble method with a bi-level beam search algorithm. The study focuses on multilingual translation in the zero-shot setting, where traditional methods like direct and pivot translations have limitations due to noise and error accumulation. EBBS synchronizes individual ensemble components through a "soft voting" mechanism, allowing for high-quality output text despite lacking parallel data. Experimental results on popular multilingual datasets show that EBBS consistently outperforms existing ensemble techniques, demonstrating its effectiveness in improving translation quality and efficiency.
Stats
Results show that EBBS consistently outperforms direct and pivot translations.
EBBS achieves higher performance in 56 out of 62 cases across two datasets.
The proposed ensemble method improves inference efficiency without sacrificing translation quality.
Quotes
"The ability of zero-shot translation emerges when we train a multilingual model with certain translation directions; the model can then directly translate in unseen directions."
"We propose EBBS, an ensemble method with a novel bi-level beam search algorithm, where each ensemble component explores its own prediction step by step at the lower level but they are synchronized by a “soft voting” mechanism at the upper level."
"Results on two popular multilingual translation datasets show that EBBS consistently outperforms direct and pivot translations as well as existing ensemble techniques."