Beyond AI in Translation: Addressing Multilingual Content Creation

Language has always played a crucial role in advertising. From the catchy brand slogan that we can all remember decades after first hearing it to persuasive and informative copy that guides customers on their purchasing journey. Language matters. Yet language is also undergoing a process of devaluation in advertising, particularly when it comes to multilingual content creation. 

It's worth noting that the word ‘translation’ itself is what’s doing language a disservice as it conjures a simplistic process that requires someone with knowledge of two languages to apply that knowledge and produce a translated text. Such language skills fall under “assumed knowledge” when the content is created by a person considered bilingual, Hannes Ben, founder and CEO of Locaria, tells us and “assumed knowledge is considered easy and is taken for granted.” Locaria is an agency specialised in global media and content activation and, naturally, Hannes knows a lot about how language functions in advertising across different markets which he incidentally explores in Locaria’s latest whitepaper.  

Hannes notes that translation is too narrow of a lens to view what actually happens when someone needs translation services. “Translation is just one element of multilingual content creation and there’s currently a big misunderstanding about what it can and can’t do.” For Hannes, this is one of the challenges of multilingual content creation; AI poses the other. “People have heard of what AI is capable of but, in the end, it’s still not enough by itself.”

Hannes wants the industry to pivot to thinking about ‘applied linguistics’ instead which encompasses “the theory of language, the theory of cultural adaptation, and the research and skills required to apply those.” By pivoting to talking about applied linguistics, the industry will start to move away from the category of assumed knowledge and its low-cost associations. “Easy concepts are cheap concepts,” quips Hannes. 

“Right now language is an afterthought in budgets.” In fact, there’s not much planning that takes language into consideration at all, according to Hannes. “With a media plan or a production plan, there’s always a strategy in place to follow. How often do you hear about a language strategy? Budgets are simply not prepared for it.”

Then there’s the literal devaluation of language services - with most translation agencies engaged in a race to the bottom in order to secure work. This means paying translators on a per word or a per batch basis - it’s barely sustainable for common languages, like English, but it’s definitely unsustainable for complex languages like Japanese and Dutch, for example. “Translation agencies are trying to survive by offering lower and lower rates which reduces the standards of the services offered. It also means that there are fewer actual language experts around because they understandably don’t want to work on very low day or word rates,” he adds. 

Another contributor to the devaluation of language services is, of course, AI - just not in the way that most picture it. AI has played a role in multilingual content creation for a very long time now. “In advertising AI has exploded in the last couple of months, but every translator will have been using a neural machine translation (‘NMT’) such as Google Translate, which is a subset of AI, to create their translation.” It’s good enough in some cases and severely lacking in others. 

“It’s good enough to use with common languages where there’s enough data but there has to be an understanding of when an expert needs to step in and fine-tune AI-assisted translation for the right audience,” Hannes explains. It’s at this point where translation ceases to be mere translation - quite basic and direct - and enters the realm of applied linguistics, meaning that a language expert calls on the tools available to them - localisation, cultural adaptation, and transcreation - to ensure that the text is best suited to the target audience. “

AI cannot (yet) speak to a client to create tailored style guides and tone of voice documents (TOV), which are subjective things. It cannot optimise and fine-tune the style of the target content based on past media performance data. It won’t come up with anything new, it will not solve a problem or help us to create a specific strategy.” It does and will continue to have its applications, however. “We have to use it for volume translations and then expert linguists come in. It’s a collaboration, a hybrid solution that allows linguists to work faster and focus on the important, creative elements of the job.”

NMT is only as good as the data it has and unsurprisingly there’s a lot of data for English, making NMT’s translation between English and other commonly used languages near perfect. It will struggle with languages where there’s less data and it will struggle with anything other than the fairly basic task of translating (e.g. transcreation and adaptation based on specific style requirements). Hannes cautions on the point of accuracy too. “You cannot blindly trust AI translations from tools such as ChatGPT which use unsupervised data which can lead to glitching and hallucinating when it doesn’t have the answer. It makes stuff up.” If the person on the other side doesn't have knowledge of the language that AI is asked to translate, it can also lead to serious mistakes as it imitates the form but not the content of what it’s translating.

What does this mean for multilingual content creation right now and for the future? “Thinking about translation only is way too restrictive,” reiterates Hannes. “We need language experts who understand where an LLM (large language model) stops, where the limits of an NMT are, and where human involvement is needed.”

“We need someone who can do for language what’s currently done with media planning. They take all the content creation methodologies available to them, the budget, the objectives of the client and any relevant performance media data points under consideration and based on that knowledge they create a plan for the client in every single market and for every asset. A translator cannot do that.” Engaging the services of an applied linguist with extensive marketing, production and media knowledge is necessary, then, to ensure that budget is spent efficiently and in areas that will make the most impact - not to fix inadequate translations or adjust badly localised strategies. “We have to focus on working with specialists, on research, and on humanising content for whatever audience we’re targeting on any channel to ensure that the content is relevant for consumers navigating a complex omnichannel journey.” AI will play a crucial helping hand but it’s on human language experts to create content that really engages.