Additive Interventions Yield Robust Multi-domain Machine Translation Models

Rippeth Elijah, Post Matt. Arxiv 2022

[Paper]    
Applications Fine Tuning Pretraining Methods Training Techniques

Additive interventions are a recently-proposed mechanism for controlling target-side attributes in neural machine translation. In contrast to tag-based approaches which manipulate the raw source sequence, interventions work by directly modulating the encoder representation of all tokens in the sequence. We examine the role of additive interventions in a large-scale multi-domain machine translation setting and compare its performance in various inference scenarios. We find that while the performance difference is small between intervention-based systems and tag-based systems when the domain label matches the test domain, intervention-based systems are robust to label error, making them an attractive choice under label uncertainty. Further, we find that the superiority of single-domain fine-tuning comes under question when training data size is scaled, contradicting previous findings.

Similar Work