Non-autoregressive Streaming Transformer for Simultaneous Translation
This work addresses latency-quality trade-offs in simultaneous translation, offering a novel architecture that could improve real-time translation systems.
The paper tackled the problem of aggressive anticipation in simultaneous machine translation by proposing a non-autoregressive streaming Transformer, which outperformed previous autoregressive baselines on various benchmarks.
Simultaneous machine translation (SiMT) models are trained to strike a balance between latency and translation quality. However, training these models to achieve high quality while maintaining low latency often leads to a tendency for aggressive anticipation. We argue that such issue stems from the autoregressive architecture upon which most existing SiMT models are built. To address those issues, we propose non-autoregressive streaming Transformer (NAST) which comprises a unidirectional encoder and a non-autoregressive decoder with intra-chunk parallelism. We enable NAST to generate the blank token or repetitive tokens to adjust its READ/WRITE strategy flexibly, and train it to maximize the non-monotonic latent alignment with an alignment-based latency loss. Experiments on various SiMT benchmarks demonstrate that NAST outperforms previous strong autoregressive SiMT baselines.