Pluggable Neural Machine Translation Models via Memory-augmented Adapters
This work addresses the problem of customizable translation for users, offering an incremental improvement over prior pluggable methods.
The paper tackles the challenge of controlling neural machine translation models to meet diverse user requirements without retraining from scratch, achieving superior performance over existing pluggable baselines in style- and domain-specific experiments.
Although neural machine translation (NMT) models perform well in the general domain, it remains rather challenging to control their generation behavior to satisfy the requirement of different users. Given the expensive training cost and the data scarcity challenge of learning a new model from scratch for each user requirement, we propose a memory-augmented adapter to steer pretrained NMT models in a pluggable manner. Specifically, we construct a multi-granular memory based on the user-provided text samples and propose a new adapter architecture to combine the model representations and the retrieved results. We also propose a training strategy using memory dropout to reduce spurious dependencies between the NMT model and the memory. We validate our approach on both style- and domain-specific experiments and the results indicate that our method can outperform several representative pluggable baselines.