One of the most visible areas AI has been in use in the area of translation. Large language models are getting better and better at learning the subtleties and nuances of human speech and becoming more accessible. Do human translators need to be worried?
Today we hear a talk on the intricacies of AI translation technology, where it is succeeding and where there is downfall, and why we’re maybe overestimating the impact it will have.
Our expert is Professor at Dublin City University and deputy director of the ADAPT Centre, Andy Way, who is a machine translation expert for over 35 years and has been instrumental in shaping the field of machine translation worldwide.
THINGS WE SPOKE ABOUT
01:12 Why neural machine translation is better than SMT
04:16 Data limitations in NMT
09:44 Hype around ChatGPT and AI
14:27 The European Language Equality Project
21:34 Inherent dangers in AI
28:42 What is ChatGPT actually being used for?
33:50 Humans are clever, not the systems
Prof. Andy Way has been in DCU since 1991, except for a period of sabbatical leave working in the translation and localisation industry in the UK between 2011-14. From 2014, he has been back in DCU full-time as Professor in the School of Computing at Dublin City University. In 2014, he became Deputy Director of the CNGL Centre for Intelligent Content at DCU. This programme was replaced by the ADAPT Centre for Digital Content Technology in 2015, where he remains Deputy Director.
Prof. Way was Editor for the journal Machine Translation from 2007-21. He was President of the International Association for Machine Translation from 2011-13, and President of the European Association for Machine Translation from 2009-15. In 2015, he received the President’s Research Award for the Sciences and Engineering faculties at DCU, and the IAMT Award of Honour in 2019 for services to the MT community.
He has over 400 peer-reviewed conference papers and journals to date, and has brought in over €60 million in external research funding.
Adapt Radio is produced by DustPod.io for the Adapt Centre
For more information about ADAPT visit www.adaptcentre.ie/
In neural machine translation, we actually do have a model of the entire source string, and because of that, that, to me, is the biggest reason why neural machine translation output is better than a statistical machine translation output. – Andy Way
The obvious implication is that for those languages where high class machine translation systems cannot be built, human translators will still be needed. – Andy Way
AI and ChatGPT is being used not only for good, but for nefarious purposes as well. – Andy Way
Large language models or multilingual large language models can produce high quality output and so, I think people who are system developers who rely on old neural technology better change to using multilingual large language models fairly quickly if their systems are not to become redundant. – Andy Way
I think there’ll be increasing demand for spoken language translation or multimodal translation in general. But again, you know, if there is a lack of data for many languages, or many use cases, for text data, you can imagine how hard this is going to be for spoken language data and multimodal data. – Andy Way
I believe that you can’t do machine translation wholly without input from linguists, or translators. – Andy Way
Maybe the honeymoon period is over, people have started to push back against tools like ChatGPT, saying that they’re not as good as people are claiming them to be and that we need legislation to make sure what use cases are being used for good rather than for evil. We need to not overhype this technology because then people are disappointed when they come to use the tools, and responsible, Explainable AI is the future. – Andy Way