Simultaneous translation, which translates sentences before they are finished, is useful in many scenarios but is notoriously difficult due to word-order differences. While the conventional seq-to-seq framework is only suitable for full-sentence translation, we propose a novel prefix-to-prefix framework for simultaneous translation that implicitly learns to anticipate in a single translation model. Within this framework, we present a very simple yet surprisingly effective "wait-k" policy trained to generate the target sentence concurrently with the source sentence, but always k words behind. Experiments show our strategy achieves low latency and reasonable quality (compared to full-sentence translation) on 4 directions: zh↔en and de↔en. * M.M. and L.H. contributed equally; L.H. conceived the main ideas (prefix-to-prefix and wait-k) and directed the project, while M.M. led the implementations on RNN and Transformer. See example videos, media reports, code, and data at https://simultrans-demo.github.io/. President Bush met with Putin in MoscowBùshí Bush zǒngtǒng President zài at Mòsīkē Moscow yǔ with Pǔjīng Putin huìwù meet prediction read write Source side → Target side → 2 Preliminaries: Full-Sentence NMT We first briefly review standard (full-sentence) neural translation to set up the notations.Regardless of the particular design of different seq-to-seq models, the encoder always takes

We propose minimum risk training for end-to-end neural machine translation. Unlike conventional maximum likelihood estimation, minimum risk training is capable of optimizing model parameters directly with respect to arbitrary evaluation metrics, which are not necessarily differentiable. Experiments show that our approach achieves significant improvements over maximum likelihood estimation on a state-of-the-art neural machine translation system across various languages pairs. Transparent to architectures, our approach can be applied to more neural networks and potentially benefit more NLP tasks.

We propose a Brillouin optical correlation-domain reflectometry (BOCDR), which can measure the distribution of strain and/or temperature along an optical fiber from a single end, by detecting spontaneous Brillouin scattering with controlling the interference of continuous lightwaves. In a pulse-based conventional Brillouin optical time-domain reflectometry (BOTDR), it is difficult in principle to achieve a spatial resolution less than 1 m, and the measurement time is as long as 5-10 minutes. On the contrary, the continuous-wave-based BOCDR can exceed the limit of 1-m resolution, and realize much faster measurement and random access to measuring positions. Spatial resolution of 40 cm was experimentally demonstrated with sampling rate of 50 Hz.

We report a novel kind of all-optical dynamic grating based on Brillouin scattering in a polarization maintaining fiber (PMF). A moving acoustic grating is generated by stimulated Brillouin scattering between writing beams in one polarization and used to reflect an orthogonally polarized reading beam at different wavelengths. The center wavelength of the grating is controllable by detuning the writing beams, and the 3 dB bandwidth of approximately 80 MHz is observed with the tunable reflectance of up to 4% in a 30 m PMF.

This paper presents a novel method that realizes simultaneous and completely discriminative measurement of strain and temperature using one piece of Panda-type polarization-maintaining fibre. Two independent optical parameters in the fiber, the Brillouin frequency shift and the birefringence, are measured by evaluating the spectrum of stimulated Brillouin scattering (SBS) and that of the dynamic acoustic grating generated in SBS to get two independent responses to strain and temperature. We found that the Brillouin frequency shift and the birefringence have the same signs for strain-dependence but opposite signs for temperature-dependence. In experiment, the birefringence in the PMF is characterized with a precision of approximately 10(-8) by detecting the diffraction spectrum of the dynamic acoustic grating. A reproducible accuracy of discriminating strain and temperature as fine as 3 micro-strains and 0.08 degrees Celsius is demonstrated.