(Wo)man Versus Machine – Part Two

Part One – How did human translators fare against their machine counterparts?

Roy Youdale of the University of Bristol shares some thoughts on the growing impact of AI on literary translation:

Master or servant? 

The field of AI and translation is changing rapidly and it is not yet clear exactly how translators, and in particular literary translators, will be affected. On the one hand, there are literary translators who argue “I think CAT tools with integrated AI are going to be the way forward, in that they enable translators to quickly accept, reject or edit AI results sentence by sentence along with ‘traditional’ machine translation, translation memories and glossary suggestions.” (comment taken from a recent Translators Association forum, October 2023), but who would strongly resist trying to post-edit complete AI-generated texts. Many translators would contend that it takes considerably longer, and is significantly more frustrating, to post-edit an AI generated translation, than it is to prepare a strong translation that’s fit for purpose from scratch.

Whose creative output trained the models?

There is also the problem that, since the large language models are trained on massive datasets including a wealth of literary translations, the past work of many translators is being used without their permission, potentially to do them out of a job!

Are we seeing a mirage?

Furthermore, because AI programs are generally trained to produce flowing prose, they can often produce superficially good translations – in that they read well and make sense – but at the expense of actually achieving an accurate rendering of the original, true to the style and tone of the piece. In other words, they make stuff up! These fake but plausible renderings are known in the AI trade as ‘hallucinations’ and for those often not conversant with the source text of a translation – think critics, publishers and readers for example – they can be almost impossible to spot.

Is AI an engaging storyteller?

At the level of texts created from scratch by AI there are also commentators such as Jennifer Foster at Dublin City University who ask the question: “Will anyone actually want to read AI-generated stories?” arguing that the output of AI tends to produce clichéd and hackneyed descriptions. In an experiment she also reports that “The first thing we found was that the AI-generated stories became more repetitive and nonsensical the longer they went on. In fact, they were so bad that there was no point even asking people to rate their quality.”

So it’s definitely a question of watch this AI space, but also “take good care”…

Recent event

Generative AI, Copyright & Publishing: Reader in Intellectual Property Law at University of Sussex, Andrés Guadamuz, sheds light on AI and copyright in the publishing industry. View recording here.

Resources

Machine Translation Literacy – open source resources to help you (to help) others evaluate the opportunities and risks of machine translation and AI in translation

Can AI help literary translators? by Roy Youdale (Goethe Institut)

Why Literary Translators should embrace Translation Technology Andy Way, Roy Youdale, Andrew Rothwell

New book on computer-assisted literary translation

For those interested in the impact of machine translation on literary translation, which has changed dramatically in the last few years with the development of neural machine translation, a new academic book deals with this in some depth, Computer-Assisted Literary Translation, edited by Andrew Rothwell, Andy Way and Roy Youdale. 

Read:

Part Three Ruth Ahmedzai Kemp explores how best to protect your copyright and contracts now that AI is here to stay