STACL: Simultaneous Translation with Implicit Anticipation and Controllable Latency using Prefix-to-Prefix Framework
By Mingbo Ma, Liang Huang, Hao Xiong, Renjie Zheng, Kaibo Liu, Baigong Zheng, Chuanqiang Zhang, Zhongjun He, Hairong Liu, Xing Li and 2 more. 2018
323 citations recorded by the source corpus.
Simultaneous translation, which translates sentences before they are finished, is use- ful in many scenarios but is notoriously dif- ficult due to word-order differences. While the conventional seq-to-seq framework is only suitable for full-sentence translation, we pro- pose a novel prefix-to-prefix framework for si- multaneous translation that implicitly learns to anticipate in a single translation model. Within this framework, we present a very sim- ple yet surprisingly effective “wait-k” policy trained to generate the target sentence concur- rently with the source sentence, but always k words behind. Experiments show our strat- egy achieves low latency and reasonable qual- ity (compared to full-sentence translation) on 4 directions: zh↔en and de↔en.
Subfields: Artificial Intelligence, Machine Translation
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Philipp Koehn
- Graham Neubig
- Felix Stahlberg
- Jiatao Gu
- Juan Pino
- Yonatan Belinkov
- Changhan Wang
- Kyunghyun Cho
In Machine Translation
Citations in the corpus
- Fairseq S2T: Fast Speech-to-Text Modeling with Fairseq (2020, methodology): “It also integrates SIMULEVAL (Ma et al., 2020a) for simultaneous ST/MT metrics such as AL (average lagging) (Ma et al., 2019) and DAL (differentiable average Lagging) (Cherry and Foster, 2019).”
- SIMULEVAL: An Evaluation Toolkit for Simultaneous Translation (2020, background): “Ma et al. (2019) define AL as AL = 1 τ(|X”
- Review of end-to-end speech synthesis technology based on deep learning (2021, background): “[133] borrowed the idea of prefix-to-prefix framework of simultaneous machine translation [132].”
- Sentence Boundary Augmentation for Neural Machine Translation Robustness (2020, unknown): “This data augmentation strategy is similar to the prefix training methods proposed in [16, 17], but we extend it to a sentence-level prefix and suffix subsampling across sentence boundaries.”
- Synchronous Speech Recognition and Speech-to-Text Translation with Interactive Decoding (2019, background): “…word can use the hidden representation of acoustic encoder, the first three transcription words, and the first predicted translation word, etc. Ma et al. (2019) proposed a wait-k policy applied in simultaneous translation where the translation decoder is always k words behind the incoming…”
- Future-Guided Incremental Transformer for Simultaneous Translation (2020, unknown): “The parameters of the incremental Transformer we proposed are exactly the same as the standard wait-k (Ma et al. 2019), while the conventional Transformer is the same as the original Transformer (Vaswani et al. 2017”
- Incremental Text-to-Speech Synthesis with Prefix-to-Prefix Framework (2019, methodology): “More specifically, the wave generation of the t th word with an inverse autoregressive flow (IAF) model can be defined as follows 2.2 Prefix-to-prefix Framework Ma et al. (2019) propose a prefix-to-prefix framework for simultaneous machine translation.”
- FINDINGS OF THE IWSLT 2020 EVALUATION CAMPAIGN (2020, methodology): “…used a hybrid pipeline for simultaneous speech 1https://github.com/pytorch/fairseq/ tree/simulastsharedtask/examples/ simultaneous_translation 3 translation track, with a Kaldi-based speech recognition cascaded with transformer-based machine translation with wait-k strategy (Ma et al., 2019).”
- Towards Incremental Transformers: An Empirical Analysis of Transformer Models for Incremental NLU (2021, methodology): “There is evidence that additional right context improve the models’ incremental performance (Baumann et al., 2011; Ma et al., 2019; Madureira and Schlangen, 2020), which results in a trade-off between providing timely output or waiting for more context to deliver more stable output.”
- Monotonic Multihead Attention (2019, methodology): “Ma et al. (2019) propose a wait-k policy where the model first reads k tokens, then alternates reads and writes.”