An often overlooked difficulty of machine translation is producing a consistent formality (or register) in the target language. This is especially hard when the source language may have fewer levels of formality than the target language. We take a transfer learning approach using Google’s AutoML Translate to train custom neural machine translation (NMT) models to consistently produce a specific formality. We experiment with formality levels for English to Spanish, English to French and English to Czech. This approach makes it possible to have better and more consistent in-context translation while still leveraging the strength of a general purpose machine translation system.
CITATION STYLE
Viswanathan, A., Wang, V., & Kononova, A. (2020). Controlling Formality and Style of Machine Translation Output Using AutoML. In Communications in Computer and Information Science (Vol. 1070 CCIS, pp. 306–313). Springer. https://doi.org/10.1007/978-3-030-46140-9_29
Mendeley helps you to discover research relevant for your work.