Universitat Rovira i Virgili

Automation and the future of translation

Masters  - 

Talk in Barcelona: "Automation and the future of translation", a càrrec d'Anthony Pym. Universitat Pompeu Fabra Dia: dimarts 13 juny del 2023 Hora: 12.45 Aula: 52.939

Neural machine translation (NMT) and large-language-based generative pre-trained transformers (GPT) present new challenges for the work of translators. By some accounts, parity with the human brain has been reached. There is little doubt that the vast majority of the words translated in the world are not produced by professional translators, and there should be even less doubt that the remaining human translations can and should benefit from various kinds of automation. Yet what this means for the training of translators is still far from clear. At what points in the translation cycle is human intervention most cost-beneficial? And if we can determine that, what are the automation-resistant skills that will be in most demand?

These questions are addressed by a study of NMT and GPT for different text types across a range of major and minor languages: English, Chinese, Spanish, French, Catalan, Russian, and Armenian. Comparison of raw outputs suggests that GPT-3 and GPT-4 do not present significant improvements on the main NMT systems, although the capacity of GPT to revise, summarize, expand and adapt texts for specific readerships should be of real use in the pre-translation phase (so we get clearer texts to translate) and the post-translation revision (so our translations are more readable).

It follows that the core translation and communication skills should still be in demand, but that translators must discover how to make automation work actively for them. Rather than bury heads in sand, we should embrace this period of turbulence as an occasion to win trust in interlingual communication, and to ensure that clients pay for our trustworthiness.

Elements relacionats


  • Twitter
  • Google+
  • Facebook
  • Linkedin