Exploring the possibilities of generative AI in (multilingual) content creation.

Generative artificial intelligence (GenAI) continues to astound us.

In January of this year when a speech in Spanish by the Argentinian president at the World Economic Forum in Davos was subsequently recreated in English, the world took note. In a video released shortly afterward and powered by generative AI, the president’s mouth was exactly synched to the English words he was saying: for all intents and purposes he became a fluent English speaker.  The video had 75 million views.

We’re becoming used to all the exciting things AI, in all its forms, can do. We marvel at the fantastical images it creates and its ability to write imaginative and entertaining stories. We are using it more and more in our everyday lives to get advice and recommendations, organize our schedules and for writing our messages and emails.

Translation is also part of its repertoire. Large language models like the GPT series and Google’s Gemini, can translate between dozens of languages and therefore potentially produce multilingual content in these languages simultaneously. There are also possibilities for speech-to-text and vice versa, with image and video creation adding yet more options.

Could generative AI be the dream solution for multilingual content creation?

The benefits of AI for creating content

These are some of the advantages that GenAI and LLMs can bring to creating content generally.

• GenAI models create content quickly and efficiently. When an LLM is ‘prompted’ by the user, its response appears in moments. Readable, grammatically correct text appears in just a few seconds.
• This efficiency makes AI-infused tasks easily scalable. And that could also apply to producing content in multiple languages.
• With the above comes cost-effectiveness. By speeding up the time it takes to write text, less money is spent on human labor.
• SEO can be boosted. AI can find the best keywords for your blog post in the blink of an eye or come up with attention-grabbing article titles and headlines within the recommended character limits.
• AI can brainstorm ideas and help with planning and strategy.

The incentives for using an AI language model to create content are therefore clear. However, in the world of AI nothing is ever that straightforward and the weaknesses of LLMs mean there are still significant drawbacks despite the enormous progress.

The downside to GenAI for content

When humans learn about a particular topic, we insert it into our already broad knowledge of the world around us. We fit it into the context of what we already know and apply our well-developed sense of logic to it.

AI systems don’t learn like us. They sift through massive amounts of data looking for patterns and their ‘logic’ predicts which words or images go together. They don’t have any experience of the world or how it functions. All they have is the data they’re ‘fed’ on.

This means that they sometimes invent their own reasoning and we can’t always trust their responses and the content they create.

• ‘Hallucination’, the invention of both fanciful or realistic-sounding responses is still an issue. As is inherent bias, when AI models produce harmful, unethical and prejudiced texts.
• AI lacks creativity and its writing style can sound generic or dull.
• There are plagiarism concerns because of how AI ‘learns’ from other sources. The New York Times has an ongoing lawsuit against OpenAI and its investor, Microsoft, claiming that OpenAI used the newspaper’s articles and content to train its AI tools.  
• Using LLMs can mean exposing any inputted data to security risks.
• Google prefers ‘helpful’, human content in its search rankings and could devalue content written by AI.

How multilingual is GenAI?

When AI is trained on texts in different languages it automatically learns how to work in those languages, although its ability is limited by how much data it consumes. This means the number of languages it ‘speaks’ can easily grow. ChatGPT is reported to be fluent in around 95 languages, Google’s Gemini over 40 and Meta’s Llama more than 30.

With the dominance of English in the digital universe, these GenAI tools could potentially offer more languages more space online, helping to promote their use and to redress the balance with the handful of languages that monopolize the internet. They could, for example, help the spread of scientific knowledge by enabling research publication in more languages or open up e-commerce opportunities.

However, the problems already present in these AI models can be exacerbated in other languages. LLMs go as far as inventing words in the languages they know less well, they tend to get confused by grammar and non-Latin alphabets, as well as mixing up different languages in the same sentences. Their best translations are usually into English and they also perform well in other dominant languages like Chinese and Spanish. When languages have few digital resources AI can learn from, it fails to translate them correctly.

AI is a handy tool

Despite its limitations, ignoring what GenAI brings to the table, especially from a business content point of view, would be unwise. Organizations are rapidly understanding what it can do and where the benefits to workflows lie. Human review continues to be a necessity but for tasks like creating product descriptions at scale, supporting the writer’s productivity or writing code, GenAI-powered models are valuable tools.

When it comes to translation, GenAI shouldn’t be used without the oversight of an expert linguist, especially for any texts where accuracy and fidelity are vital, but that doesn’t mean it’s not helpful. For non-commercial purposes it can be used in a similar way to other available machine translation tools, for example, if you want the gist of a text in another language or to compose a simple message.

The language industry is now integrating AI technology into its translation processes. CAT (computer aided translation) and TMS (translation management system) tools customized with AI, are leading to extra efficiency and productivity. Language service providers are skilled at recognizing the advantages of the latest technologies and quickly adopting them for time and cost efficiencies. AI is no exception.

The road ahead for GenAI

As we begin to understand more about GenAI and what it does best, the more we are able to use it in the most effective ways.

AI for translation is no exception. In Osaka, Japan, a multilingual chatbot that uses GenAI, is helping tourists navigate the city; ancient manuscripts in extinct languages are being translated with the help of AI to unlock the secrets of past societies; and media outlets like the BBC are using AI to help diffuse their news articles in more languages.

In all these cases, the work of human experts and linguists is augmented by AI.

Creating content, whether in multiple languages or not, has much to gain from GenAI and it will undoubtedly help the content ‘explosion’ that we are currently experiencing. But when you need to be sure the text is absolutely fit for purpose, trust human professionals to help you reap the rewards GenAI can bring.

Talk to t’works today about how we use AI to benefit your language projects.