KIT | KIT-Bibliothek | Impressum | Datenschutz

Towards Stream Translation: Adaptive Computation Time for Simultaneous Machine Translation

Schneider, Felix; Waibel, Alexander

Abstract (englisch):

Simultaneous machine translation systems rely on a policy to schedule read and write operations in order to begin translating a source sentence before it is complete. In this paper, we demonstrate the use of Adaptive Computation Time (ACT) as an adaptive, learned policy for simultaneous machine translation using the transformer model and as a more numerically stable alternative to Monotonic Infinite Lookback Attention (MILk). We achieve state-of-the-art results in terms of latency-quality tradeoffs. We also propose a method to use our model on unsegmented input, i.e. without sentence boundaries, simulating the condition of translating output from automatic speech recognition. We present first benchmark results on this task.


Postprint §
DOI: 10.5445/IR/1000120793
Cover der Publikation
Zugehörige Institution(en) am KIT Institut für Anthropomatik und Robotik (IAR)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2020
Sprache Englisch
Identifikator KITopen-ID: 1000120793
Erschienen in Proceedings of the 17th International Conference on Spoken Language Translation. Ed.: A. Waibel
Verlag Association for Computational Linguistics (ACL)
Seiten 228–236
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page