For ages, there has been an unspoken consensus that when creating UX content, you need to adapt your writing for localization. This convention was born out of necessity, understanding that localization is not a simple linear process. It involves a network of interconnected tasks, each with their own set of intricacies. Considering the numerous variables and complexities involved in localization, it makes sense to simplify the source. Uncertainties in the original content could exacerbate and cause significant disruptions down the line.

Let's consider a UX copy created for an English-speaking audience, featuring idioms or culturally-specific references. If this copy were to fall into the hands of someone not so familiar with the source culture, it might be localized without any adaptation. And it may certainly be confusing for a non-English-speaking user. An unadapted phrase like "raining cats and dogs" might lead a Spanish user to expect literal felines and canines, rather than a downpour. It's clear that any foggy areas in the source content could snowball into significant issues when transferred to a different cultural context.

Therefore, it becomes clear why we've adhered to the practice of "writing for localization" in creating UX content. By taking the time to adapt the copy, says the industry, we can streamline the localization process, mitigating the chances of misunderstanding or misinterpretation. We also help to ensure a more user-friendly experience in every market where the product is available, affirming the global nature of our digital age.

Should we write for localization?

But the mandate to "write for localization" led us to strip our copy of all its personality, resulting in clear but stark and lifeless writing. We removed any cultural references, nuances, or distinctive voices to ensure absolute clarity. The result? Text that was precise but as engaging as a tax form.

This goes against the trend of the past few years, where conversational, engaging language has become the gold standard for UX copy. We've migrated from stiff, formal language to a more conversational, engaging approach. This shift reflects our increasing understanding of the user's need for intuitive, interactive, and human-like digital experiences. Think of the messages that pop up when you're filling out an online form, the words on a button, or the subtle instructions that help you navigate an app. Today's microcopy is often personable, aiming to create a friendlier user experience. Gone are the days of "Invalid input"; now we're more likely to see "That doesn't look right."

However, when we put this vibrant, engaging UX copy under the lens of localization, the picture gets a bit complicated. The cultural context, idioms, and informal language that make copy engaging also make it tricky to localize. While maintaining the clarity and accuracy of our copy is vital to avoid any localization-related issues, there's a growing need to keep our UX copy engaging and inspiring for users, aligning with the original source as much as possible. It’s a balancing act between maintaining the spark of the original and ensuring clear understanding across multiple languages and cultures. The challenge we face now is adapting to this new gold standard while maintaining the clarity needed for successful localization.

Can we do it differently?

There is an alternative method to achieve this - providing detailed guidelines for each string to the linguists, outlining the context, and the key messages we want to convey. Instead of leaving the translation team to guess the meaning or the importance of a certain element, these guidelines provide clear-cut instructions.

Imagine you are localizing a banking app. In its source form, the app uses professional yet approachable language, with clear instructions and perhaps a touch of finance jargon. It is intended to project trust, security, and user-friendliness. Let's take a simple example: the copy for a button that initiates a money transfer might read, "Swipe to send money securely."

In this case, the string isn't just instructing the user to swipe; it's also reinforcing the app's focus on security. By providing linguists with context and intent for this string, we ensure they understand the dual purpose of the text. They'll know that the localized copy should not only instruct users about the swipe action but also reassure them about the security of their transaction.

But here's the catch: The level of detail required in these guidelines demands a substantial time investment. For every piece of copy, there needs to be an accompanying set of instructions. Writing these guidelines is a painstaking process, demanding a deep understanding of the product, the brand voice, the intended user response, and the specific nuances of the source language. Additionally, these guidelines need to be clear, precise, and easy for the translator to understand.

The time drain doesn't stop at writing the guidelines. They also lengthen the translation process. The linguists now have an additional document or set of texts to study and comprehend before they can even begin translating.

Enter: technology

The advent of AI and LLMs (Large Language Models) has transformed the landscape of machine translation. These groundbreaking technologies have expanded our toolbox, adding new capabilities that can be game-changers in the localization process.

AI and LLMs can be likened to quick learners. They are capable of receiving complex, detailed instructions and processing them at a rapid pace, outmatching any human capacity for speed. This ability makes them particularly useful in the context of localization. They can absorb and interpret extensive guidelines for each string of text, thereby aligning the localized output with our preset criteria.

Let's revisit our banking app example. Instead of a human translator pouring over pages of instructions, we could provide these guidelines to an LLM. The LLM could quickly analyze these directives, comprehend the context and intent, and try to accurately translate the UX copy accordingly. Even complex tasks, such as understanding the dual function of the 'Swipe to send money securely' button, may very well be within the capabilities of advanced LLMs in the near future.

Unlike traditional Machine Translation (MT), which is limited by rigid algorithms and fixed phrase databases, LLMs are designed to be adaptable. They can handle nuances and generate unique solutions tailored to specific cultures when provided with ample multilingual data.

Can AI translate alone?

While LLMs and AI bring numerous benefits to the localization process, like any technology, they're not without their problems. One such problem is their tendency to "hallucinate." They sometimes generate content that, while plausible, wasn't originally included in the source text or the provided guidelines.

Returning to our banking app example, an LLM might take a phrase like "Swipe to send money securely" and interpret security in a broader sense. Consequently, it might generate a translation that implies additional safety features, like fraud protection or insurance, that the app doesn't actually offer. These extraneous details, while well-intentioned, could mislead users and create unrealistic expectations.

This is where the role of detailed instructions becomes crucial. It's akin to providing our LLM with a map that clearly outlines the boundaries of its creativity. These instructions not only guide the LLM in understanding the context and intent of the original copy, but they also set clear limits to prevent the addition of unwarranted content.

Moreover, despite the impressive capabilities of AI and LLMs, the human touch remains an invaluable part of the localization process. A proficient human translator or a localization expert should always review the LLM-generated content before it goes live. They can spot and correct any deviations or hallucinations, ensuring that the localized copy stays true to the original message and intent.

The secret is in the combination

The interplay between advanced LLMs, detailed instructions, and human involvement brings us to a promising juncture in the localization journey. It allows us to balance speed and accuracy, maintain the engaging qualities of UX copy, and prevent misunderstandings due to overzealous AI translations. It's an exciting step towards redefining what it means to write for localization.

This synergy between LLMs and detailed instructions can create a multilingual UX copy that is more fluent and less robotic than what we would usually get. Additionally, LLMs can expedite the creation of these instructions. We can feed the AI a brief summary or even a recorded explanation, and it will efficiently write the guidelines for us.

All these advancements prompt a reconsideration of our practices when it comes to "writing for localization". We don't have to make our copy characterless anymore; instead, we can start articulating our intended meanings in a more engaging way.

As we ride this wave of change, let's embrace this evolution in writing for localization. This technology is new, but it's growing and evolving at an extremely fast pace. Our goal should be to keep our content lively and clear, make sure we're explaining our intentions, and, yes, also use technology to make the process more efficient. With this mindset, we can change localization from a task into an opportunity to improve the experience for our users.