Neural Simultaneous Speech Translation Using Alignment-Based Chunking

Proceedings of the 17th International Conference on Spoken Language Translation (IWSLT), pages 237–246 July 9-10, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 237 Neural Simultaneous Speech Translation Using Alignment-Based Chunking Patrick Wilken, Tamer Alkhouli, Evgeny Matusov, Pavel Golik Applications Technology (AppTek), Aachen, Germany {pwilken,talkhouli,ematusov,pgolik}@apptek.com Abstract In simultaneous machine translation, the objec- tive is to determine when to produce a partial translation given a continuous stream of source words, with a trade-off between latency and quality. We propose a neural machine trans- lation (NMT) model that makes dynamic deci- sions when to continue feeding on input or gen- erate output words. The model is composed of two main components: one to dynamically decide on ending a source chunk, and another that translates the consumed chunk. We train the components jointly and in a manner consis- tent with the inference conditions. To generate chunked training data, we propose a method that utilizes word alignment while also pre- serving enough context. We compare models with bidirectional and unidirectional encoders of different depths, both on real speech and text input. Our results on the IWSLT1 2020 English-to-German task outperform a wait-k baseline by 2.6 to 3.7% BLEU absolute. 1 Introduction Simultaneous machine translation is the task of generating partial translations before observing the entire source sentence. The task fits scenarios such as live captioning and speech-to-speech transla- tion, where the user expects a translation before the speaker finishes the sentence. Simultaneous MT has to balance between latency and translation quality. If more input is consumed before transla- tion, quality is likely to improve due to increased context, but latency also increases. On the other hand, consuming limited input decreases latency, but degrades quality. There have been several approaches to solve si- multaneous machine translation. In (Dalvi et al., 2018; Ma et al., 2019), a fixed policy is introduced 1The International Conference on Spoken Language Trans- lation, http://iwslt.org. to delay translation by a fixed number of words. Alternatively, Satija and Pineau (2016); Gu et al. (2017); Alinejad et al. (2018) use reinforcement learning to learn a dynamic policy to determine whether to read or output words. Cho and Esipova (2016) adapt the decoding algorithm without re- lying on additional components. However, these methods do not modify the training of the under- lying NMT model. Instead, it is trained on full sentences. Arivazhagan et al. (2019) introduce a holistic framework that relaxes the hard notion of read/write decisions at training time, allowing it to be trained jointly with the rest of the NMT model. In this paper, we integrate a source chunk bound- ary detection component into a bidirectional recur- rent NMT model. This component corresponds to segmentation or read/write decisions in the litera- ture. It is however trained jointly with the rest of the NMT model. We propose an algorithm to chunk the training data based on automatically learned word alignment. The chunk boundaries are used as a training signal along with the parallel corpus. The main contributions of this work are as follows: • …

Content truncated to 500 words