How Transferable are Attribute Controllers on Pretrained Multilingual Translation Models?

TL;DR

This paper explores the transferability of attribute controllers in pretrained multilingual translation models, demonstrating effective zero-shot attribute control without additional supervised data, especially for low-resource languages.

Contribution

It introduces a gradient-based inference-time controller for attribute control that transfers well across languages and complements finetuning, reducing the need for attribute-annotated data.

Findings

Inference-time control performs well in zero-shot settings.

Controller transfers effectively to distant languages.

Combines well with finetuning for improved control.

Abstract

Customizing machine translation models to comply with desired attributes (e.g., formality or grammatical gender) is a well-studied topic. However, most current approaches rely on (semi-)supervised data with attribute annotations. This data scarcity bottlenecks democratizing such customization possibilities to a wider range of languages, particularly lower-resource ones. This gap is out of sync with recent progress in pretrained massively multilingual translation models. In response, we transfer the attribute controlling capabilities to languages without attribute-annotated data with an NLLB-200 model as a foundation. Inspired by techniques from controllable generation, we employ a gradient-based inference-time controller to steer the pretrained model. The controller transfers well to zero-shot conditions, as it operates on pretrained multilingual representations and is attribute -- rather than language-specific. With a comprehensive comparison to finetuning-based control, we demonstrate that, despite finetuning's clear dominance in supervised settings, the gap to inference-time control closes when moving to zero-shot conditions, especially with new and distant target languages. The latter also shows stronger domain robustness. We further show that our inference-time control complements finetuning. A human evaluation on a real low-resource language, Bengali, confirms our findings. Our code is https://github.com/dannigt/attribute-controller-transfer