AI Chatbots? Fun Toys, But How Can They Help me in Localization?

By Dr. Rafał Jaworski, AI Engineering Executive, Tech Lead at PwC

Recently, in a supermarket, I saw a big-screen TV displaying footage from a car race. Being a true motorsports fan, I immediately recognized the Monza circuit, car makes, and models, but I struggled to recognize any of the drivers. Only after a prolonged stare (which took longer than I would like to admit) did I realize that it was not actual footage but AI-generated

content. It is even easier to be deceived by conversational AI - it really feels like talking to a human. But here’s the question that has to be asked: How are those impressive AI capabilities going to help humans? Specifically, what can AI do for localization?

One thing that comes to our minds is to ask the AI chatbot to translate something for us. After all, modern LLMs have been trained on vast amounts of text in multiple languages. Thanks to their knowledge modeling abilities, they understand words, phrases, sentences, concepts, and meanings - all in different languages. What’s worth noting is that they are the first automatic mechanisms to understand the pragmatic side of language, such as hidden intent or irony.

Wonderful! We now have automated polyglots that we can politely ask to translate,

proofread, or rewrite content for us. All that with the touch of a magic wand - the prompt. Send a prompt and get an answer like from an ancient oracle - as simple as that.

But here’s a small problem - can we really let AI loose? Can we publish AI-translated or generated material without checking? This concern has moved the focus of AI research to AI Agents. As opposed to chatbots, agents use various tools and data sources to perform actions and justify their decisions. AI agents are perfect tools to assist humans in their work. Using an agent, a human translator can receive actual help - suggestions for translations (backed by examples from translation memory), rewriting alternatives, or mistranslation warnings.

It is extremely important to keep the human-in-the-loop model to achieve quality and efficiency in localization processes.