A Use Case of LLMs in Transcreation

I have already discussed on social media and in other blog posts the reasons why I prefer not to rely on machine translation or AI to produce my translations. Despite major improvements in recent years, the results still do not meet my standards, and the many adjustments I make to improve them tend to slow me down rather than enhance my productivity. However, LLMs (large language models) offer other interesting applications that can assist meticulous linguists like me who are always keen to refine their knowledge and skills.

One of the most enjoyable features of LLMs is their ability to interact in natural language, often providing enlightening feedback and stimulating new ideas. ChatGPT has a particularly charming way of conversing, while Claude is proactive, making supplementary suggestions and encouraging further exploration. For solopreneurs like me, such a virtual team can provide valuable assistance, especially when I need to generate ideas for the creative aspects of my work.

Recently, I experimented with ChatGPT for the transcreation of adlobs in the pharmaceutical field.

An Adequate Strategy is Key

In my initial attempt, I simply asked ChatGPT to suggest three “creative translations” for each of the slogans I entered. Unsurprisingly, this approach fell flat, as ChatGPT’s suggestions were literal translations that offered only slight variations on the original wording. Transcreation, however, involves more than rephrasing in the target language: it’s about conveying emotions, addressing specific concerns, and connecting with people on a deeper level. Since different cultures respond to different stimuli, certain words or contexts can elicit very different reactions. Transcreation requires adapting a message for a target audience and sometimes involves significant creative changes. As this approach wasn’t achieving what I needed, I soon abandoned this first collaboration attempt with my virtual assistant and resumed ‘manually’.

For my next transcreation task, a different strategy came to my mind: rather than seeking fully-formed translations, I focused on using ChatGPT to generate ideas and explore variations on words and concepts. This approach proved far more productive and satisfying than my previous attempt. Although there’s no single ‘right’ method, and my approach is certainly open to improvement, I hope the process I outline here provides useful inspiration if you’re new to LLMs or looking for ways to use them.

Transcreating with ChatGPT

Setting the Framework

I started by telling ChatGPT I wanted to collaborate on transcreating a series of adlobs for a vaccine campaign. ChatGPT welcomed the idea, asking questions to gather the campaign’s key elements: core message, tone of voice, target demographic, and the visual elements in the ads. I provided details on these elements, adding a few other points that ChatGPT hadn’t asked about but omitting the specific visuals for each adlob at this stage, as I planned to work on them separately.

ChatGPT then suggested a method for the process, which was relevant and an interesting reading, but not what I wanted at that point.

Brainstorming to Shape Ideas

I invited then ChatGPT to brainstorm around specific ad components. For example:

• The English word "VAX" appeared in a small graphical element of an image. In French, there’s no such abbreviated term for "vacciner/vaccination." I asked ChatGPT to suggest alternative short words or concepts associated with vaccination that might work in the same context while conveying a positive image of vaccines.
• I asked for positive concepts associated with the target audience (people over 50) to identify emotions that I could later incorporate into the slogans or body text.
• For each ad, I described the image, text, and message to be conveyed, then asked ChatGPT to suggest slogans that captured the desired emotions.

Some of ChatGPT’s suggestions were immediately usable, others provided ideas or concepts that I could combine to create a strong slogan, while others simply didn’t fit. Regardless, ChatGPT provided a useful base to refine and build on. I re-submitted some of my reflections to ChatGPT for further brainstorming and alternative phrasing.

Refining the Style

There are times when even linguists have a word “on the tip of the tongue”, and thesauri don’t always provide the spark needed. When deadlines are tight, LLMs can be helpful.

If I had a solid sentence or slogan but wanted a stronger word to convey a specific emotion, I’d ask ChatGPT to suggest alternatives, explaining what I wanted to express. I could then choose the best option from several suggestions.

For structural refinements –like shortening sentences, suppressing repetitions, or lightening the style– I’d request multiple rephrasings from ChatGPT, giving clear instructions about the results to be obtained. Requesting several options is beneficial, as only one of the suggestions may be suitable, or you might combine elements from different suggestions for an optimal solution.

If you’re pleased with the idea you generated for a slogan but want variations on it, you can ask the LLM to generate alternatives, which might lead to more refined ideas or spark more ideas to use across a campaign’s adlob series. Here also, it is advisable to state precisely the emotion or idea you’re aiming to express.

Finally, you can also simply ask the LLM to improve the style of your draft, then select only the elements that truly enhance your original text. Don’t yet omit to review the generated suggestions critically, as they may introduce subtle errors or deviate from the intended message.

Back-Translation and Proofreading the Rationales

One straightforward use for LLMs in transcreation is back-translation, which allows you to give clients a sense of how you’ve adapted their message. A back-translation doesn’t need to be perfect: it only has to convey the choices made during the transcreation process. LLMs can translate your work fairly literally, saving valuable time, though you should still proofread to ensure accuracy in key details.

You’ll need to write the rationales for your creative choices yourself, but an LLM can proofread them to catch grammatical errors or help ensure that your explanations are clear and accessible to your client.

Conclusion

As with all tasks involving LLMs (including translation!), LLMs don’t replace your creativity, intelligence, or critical thinking to govern and guide the process properly. But when used strategically, they can significantly boost productivity, creativity, and final quality.

Transcreation is a challenging task involving cultural and linguistic considerations. Depending on the topic or your mood, searching for creative ideas and impactful language can consume hours. Working in dialogue with ChatGPT allowed me to challenge my initial ideas, broaden my perspective, and explore a range of alternative wordings while sparing me the frequent stress of overrunning due to the complexities of word choice and idea development. It enabled me to work not only more efficiently but also more enjoyably, completing the task comfortably within the allotted hours and receiving enthusiastic feedback from a satisfied client.

Visit my Website