Translate Smart, not Hard: Cascaded Translation Systems with Quality-Aware Deferral

TL;DR

This paper introduces a cascaded machine translation system that uses quality estimation metrics to selectively defer difficult instances to larger models, achieving high performance with reduced computational costs.

Contribution

It proposes a simple, QE-based deferral method for cascaded translation systems, improving efficiency while maintaining translation quality.

Findings

QE-based deferral matches larger model performance

Reduces model invocation to 30-50% of cases

Validated through automatic and human evaluations

Abstract

Larger models often outperform smaller ones but come with high computational costs. Cascading offers a potential solution. By default, it uses smaller models and defers only some instances to larger, more powerful models. However, designing effective deferral rules remains a challenge. In this paper, we propose a simple yet effective approach for machine translation, using existing quality estimation (QE) metrics as deferral rules. We show that QE-based deferral allows a cascaded system to match the performance of a larger model while invoking it for a small fraction (30% to 50%) of the examples, significantly reducing computational costs. We validate this approach through both automatic and human evaluation.