Large Language Models and Translation: Should MT Developers and Human Translators be Worried

Professor Andy Way speaks to the Chinese University of Hong Kong, Shenzhen, China.


With the paradigm shift to neural machine translation, we have seen a huge improvement in quality.  However, this has led to a level of overhyping of MT capability, which has led some commentators to invoke ‘the singularity’ and claim that human translators are doomed as their jobs will be taken over completely by machines.  While performance gains using AI techniques are impressive, they do not merit claims from Meta CEO Mark Zucherberg who recently stated that “the ability to communicate with anyone in any language: that’s a superpower people have dreamed of forever, and AI is going to deliver that within our lifetimes”.  Whenever there is a paradigm shift, performance gains are seen quite quickly, but level off reasonably soon thereafter.  Based on 35 years of building MT systems of different types, this talk attempts to apply some perspective as to what the future holds, especially for human translators, who I content will continue to play a critical role in further performance gains are to be seen, and MT is to be more widely adopted.

Full recording.