Generating Diverse Translation by Manipulating Multi-Head Attention
This addresses the need for diverse outputs in machine translation and related tasks, offering a method to enhance data augmentation, though it appears incremental as it builds on existing Transformer attention mechanisms.
The paper tackles the problem of generating diverse translations by discovering that different attention heads in Transformer models align to different translation candidates, and proposes manipulating these heads to produce varied outputs while maintaining quality. Experiments show the method generates diverse translations without severe quality drops and improves translation performance when used with back-translation for data augmentation.
Transformer model has been widely used on machine translation tasks and obtained state-of-the-art results. In this paper, we report an interesting phenomenon in its encoder-decoder multi-head attention: different attention heads of the final decoder layer align to different word translation candidates. We empirically verify this discovery and propose a method to generate diverse translations by manipulating heads. Furthermore, we make use of these diverse translations with the back-translation technique for better data augmentation. Experiment results show that our method generates diverse translations without severe drop in translation quality. Experiments also show that back-translation with these diverse translations could bring significant improvement on performance on translation tasks. An auxiliary experiment of conversation response generation task proves the effect of diversity as well.