In 2023, the translation industry is a unique blend of specialised human work and applied technology at every level.
The lowest level comprises machine translation with 99.99% of global translations. Machines translate over two trillion words a day in apps, information monitoring systems, and browsers. Yet they represent only about 2% of the translation industry’s turnover.
At the centre is professional human translation, which comprises over 90% of expenses while processing less than 0.01% of word volume. And over two-thirds of human translation is the post-editing of machine translation. The global share of machine-assisted translation is growing.
The top dog in the translation world is the creative translation of slogans, marcom, headlines, and other artistic content where emotions pack a punch. Creative translation involves the re-creation of a text in another language, or even the use of completely new phrases. This service remains more expensive and time-consuming compared to standard translation.
Let’s take a look at how ChatGPT will affect each of these three levels.
ChatGPT is a creative translation game-changer
It’s right here that ChatGPT will have the most impact. AI can also compose actionable slogans, poems, and songs, and even create instruction-led puns. Writers and students already use ChatGPT to generate texts, which they then edit like MTPE.
Companies are taking turns to see what ChatGPT can do. During this testing period, companies will decide to what extent they can integrate ChatGPT into their current processes. Several companies have announced that ChatGPT has enabled them to increase the speed and reduce the cost of content creation, and are retraining people to check outputs and verify facts.
Creative translation (transcreation) used to be a manual task with Microsoft Word or Google Docs. Now generative models like ChatGPT can be part of these tools, and work on the principle of the autosuggest function.
Creative translators will be able to quickly generate much more content through inputs into the tool and by checking outputs. Increased productivity means lower costs and fewer manual tasks for translators.
As with every new technology, there will be initial push-back. Some institutions will ban AI-based copywriting. For example, the ICML conference has forbidden ChatGPT-written content in academia. In the social division into technocrats and Luddites , creative translation and copywriting will be divided into “manual” and “machine” categories, similar to translation and post-editing, with a significant increase in volume in the latter category.
Conclusion: ChatGPT will fundamentally change copywriting and creative translation, with a tsunami in word volumes.
How ChatGPT affects professional translation
Compared to creative translation, the impact on professional translation will be less pronounced. Rather than a revolutionary sea change, it will initiate a permanent innovation that will incrementally necessitate less human post-editing. Fact-checks and translation revisions will still be required for professional texts, even with support from ChatGPT and other generative models. There will certainly also be push-back from individuals and companies to fully rely on these models.
API will enable GPT-4 to be variously embedded in CAT tools and TMS. GPT-4 can help with:
- text quality optimisation,
- editing of related matches.
This will reduce the word volume that translators have to check, and so increase the number of words they can revise per hour. Imagine Smartling, Phrase, XTM, memoQ and RWS Trados all integrating GPT in some way. 😊
A translator can currently translate an average of 450 – 900 words per hour. Within two years, the average translator may be able to process 1,500 – 2,000 words per hour with the latest AI.
Conclusion: AI will increase translators' productivity. But translation agencies will still have a key role in this ever-evolving linguistic landscape.
Development of machine translation and post-editing in an AI world
By ‘machine translation’ we mean the service provided by a translation agency. Professional linguistic teams and supportive software and plugins ensure subsequent text editing, thereby increasing the quality of machine translation.
MT accuracy is then sufficient for customer support, technical documents, internal comms, product descriptions in e-shops, e-learning courses, and other basic documentation.
There are several potential use cases of large language models in this area, ranging from identifying errors in translations to rewriting low-quality source text to facilitate its translation.
ChatGPT already translates some language combinations better than DeepL and Google Translate .
Yet ChatGPT doesn’t currently pose a threat to machine translation tools, or companies providing translation and localization services. Jochen Hummel, the creator of Trados, also announced this. Hummel does though point to a future where chatbots will revolutionise translation workflows. If ChatGPT can generate and rewrite content, why not use it for post-editing machine translation too?
What are the advantages of large language models compared to traditional neural MT systems?
Their flexibility allows a user to change the tone and adapt translations for various audiences. For example, instructions such as “translate text suitable for a 5-year-old” or “remove gender bias from the following translation <…>”.
What are the main drawbacks of using ChatGPT for MT?
Infrastructure costs. Period. While an MT model can run on a single CPU or GPU, running a large language model requires an expensive multi-GPU configuration. Only large tech companies will be able to afford such scale.
And such a transition will burden even larger organisations. It may take companies years to switch to machine translation via ChatGPT, due to concerns about security and their past investments in neural machine translation systems. So even though we can translate using ChatGPT, this will be held back.
Note by article’s writer:
Are our ChatGPT conversations private?
No. OpenAI explained that as part of its commitment to safe and responsible AI, it examines our conversations to improve its systems and to ensure that content conforms with its security policy. So it’s not recommended to disclose any personal or private information to ChatGPT.
Conclusion: Large language models (like ChatGPT) represent the next generation of MT technology. But since the translation industry is an established and professional market, change will not be as fast or far-reaching as for copywriting.
How should you respond if you own a translation/localization agency in 2023? Start pilot projects with GPT in content creation and transcreation, where the impact has already been immediate. Much will change by year-end.