The central claim of Martin Kay’s famous article “The Proper Place of Men and Machines in Language Translation” is that one cannot automate what one does not fully understand – in this case, translation. In light of the impressive quality of the translations produced by current neural machine translation systems (henceforth, NMT), we return to that claim and examine whether it still obtains for NMT, which implements a variety of distributional semantics. Acknowledging the obvious, i.e. that machines clearly do not understand in the same way that humans do, we contend that NMT translations are indeed meaning-based. In support of that contention, we point to the success of certain NMT systems in achieving so-called zero-shot translation between languages where no explicit training data are available to them. We conclude by suggesting that these systems now require us to invert the proper roles of men and machines in language translation.

Enregistrement de la présentation : https://drive.google.com/file/d/19TSGM6rzXNMI8c5buwMkhDqYEikY--jL/view?usp=drive_link