Translation comparisons for IC using AI and other software

After the English copy was approved, I needed to include a French-translated copy in the email. I'd heard that Ai gen translations were improving, and in particular, I'd read ChatGPT was doing swell. I decided to run a translation challenge between ChatGPT, Deepl and Google Translate.

Working in IC for as long as I have and at various global organisations, the feedback regarding machine-translated comms was very poor. Deepl was the closest I'd seen that needed less editing by a native speaker. But, it still needed a person to pick up the nuances, style and tone. This was going to be a good challenge.

The image below shows the results. I used ChatGPT to translate my English version. I then sent this translated version to my French-speaking colleague to check. They came back, of course, with a few edits. I decided to highlight the differences between ChatGPT and my colleague's version. I then ran the translation again with Deepl and Google Translate to compare it with ChatGPT's copy. The results...

ChatGPT came out on top against Deepl and Google Translate, which both had many more incorrect translations. It was good to see the comparison to build up my trust and knowledge. Also, there's an opportunity to train ChatGPT to look at the comparisons and spot any inaccuracies, which I'll be doing with it (of course removing any real-world identifiers).

The key differences I could see had to do with the structure of certain paragraphs and the choice of what to lead with regarding the thing it describes. As well as how a real person would write (based on their experience and understanding of what you are really trying to say versus the predictive approach by these automated tools.)

Ai can't truly understand the nuances of what is being said and appreciate the audience, tone, style, feelings and emotions. The automation also can't appreciate that translation is not always about an exact translation of written English but a translation of the sentiment; therefore, a person would phrase and structure it differently.

Some nice insights here for all comms folk to think about. It's easy/lazy to use future AI platforms to give us a machine translation in the comms, which most native speakers/readers can tell a mile away. Spend the time to ask someone to review before just sending comms out if you truly appreciate your audience to want to engage with your written word.