AI translation tools – at the moment at least – cannot produce 100% accurate translations. The current hype surrounding GPT-4 and generative AI tools strongly keeps on, but there is certainly room for improvement in order for the technology and for like tools to tackle the complexities of accurate translation.
That being said, not every type of language project requires 100% translation accuracy. As strange as it might sound, there are projects where you would have to prioritise creativity over accuracy – and this is where the next breed of AI translation tools could make their biggest impact.
What do we mean by creative translation?
Creative translation broadly refers to translation services that handle highly-creative content. This can include marketing and advertising content, fictional literature or anything that includes creative language – the kind that is typically lost in direct translation.
If you want a more in-depth explanation, take a look at this article: All You Need to Know About Creative Translation Services.
Here’s a quick snippet of how we define creative translation in that article:
“Creative translation refers to a specialist set of language services designed to overcome the challenges of translating creative messaging. In general, the more creative the original message is, the higher of a chance it has of getting lost in direct translation because creative language inherently deviates from standard meaning and usage.”
For example, slogans can use a lot of wordplay and indirect meaning. Writers use metaphors and other literary devices to convey messages. Even public speakers use a range of creative devices to get their points across and make them more memorable.
The problem is that these creative elements tend to lose their meaning when translated directly. They either make no sense at all or – at best – the impact you have worked so hard to create with your original message, loses its strength.
To overcome the challenges of translating creative content, in most cases, you would have to take a less direct approach. Instead of translating the source content with 100% accuracy, your focus should lay on the intended meaning or the impact – and capture this in the most appropriate way for each target language.
In some cases, you can end up with a completely different message in the translated version, but that’s okay as it’s part of the process.
For example, directly translating the English expression “I wash my hands of you” into most languages would make no sense at all. So, you would need to find the equivalent metaphor (provided this exists in your target language) or a similar one that is leaning more towards the idea of no longer taking responsibility or caring about someone.
When translating this expression directly into Spanish, there is a problem. The equivalent phrase in Spanish also has a metaphorical meaning, but it is closer to the English expression of “passing the buck” or placing the blame on someone else. So, in this case, the direct translation makes more sense, but it means something completely different.
To overcome this and similar translation issues, you need to find a completely different way to capture the intended meaning of your message – and that could often be with a completely new message for your target language(s).
As things stand, AI translation tools are only capable of handling the most basic translation tasks with acceptable levels of accuracy. Google Translate can do a decent job of translating phrases like “How are you?” into a variety of languages, but it quickly runs into problems with complete sentences, conversations and pieces of content.
As we have seen, GPT-4 and the latest breed of generative AI tools outperform Google Translate with some translation tasks. However, Google Translate outperforms them at other tasks and, overall, the quality of output is around the same level.
What if we start using AI generative tools for translation tasks that don’t require 100% accuracy?
Let’s say we are translating a company slogan and capturing a similar meaning is proving particularly difficult. Standard machine translation tools like Google Translate can only translate the text we input. However, with generative tools like GPT-4, we can use broader prompts and provide feedback to explore ideas around meaning, which is very useful.
You can ask GPT-4 to translate a slogan into your target language and see what it comes up with. Next, you can say, “make it sound happier” or “translate it for kids” to generate more ideas. In addition, you can add new phrases and adjectives like “delicious” and ask it to incorporate them or make the slogan and your message sound more natural in everyday language.
As the technology continues to improve, language experts will be able to use and experiment with generative AI tools for translation, to explore such ideas and concepts – and this is the area where GPT-4 for example differs from Google Translate. So, with AI translation, we are now reaching the point where algorithms can help a lot more with the more creative aspects of language services.
At translate plus, we are very excited about the future of generative AI tools and we have been experimenting with generative AI technology for quite some time now. We are also currently testing Whisper’s impressive capabilities and speed to produce solid output for speech-to-text transcriptions. Our current investigation and testing mainly gravitate around the application of these tools as alternatives to machine translation with post-editing (MTPE) and to human transcription for faster delivery of multimedia translation projects requiring subtitling, transcription and voiceovers.
If you are finding it difficult to translate creative content like advertising copy, slogans and marketing messages, the creative language services department at translate plus can help. Call us on +44(0)20 7324 0950 or fill out the form on our contact page and one of our creative language specialists will get back to you.