Is There Such a Thing as Too Much Innovation in Localization?

Technology in the localization industry continues to quickly evolve, offering businesses faster, higher-quality, and more cost-effective ways to bring their global products to market. There are now numerous options to optimize translation delivery, including cloud platforms, CAT tools, TM tools, APIs, quality checkers, and multilingual AI.

Yet, most professionals agree that humans must always remain in the loop to make sure goals are met, brands are protected, and customers get content with the correct cultural nuance. As process innovation often leads to increased automation, where should we draw the line to prevent over-automating processes where human supervision remains important?

What are the potential ethical concerns and organizational risks of relying too heavily on automation? How can we integrate innovation into our organizational culture responsibly and ethically? Striking this important balance is an open question. While we don’t aim to provide all the answers, we offer our perspective from 20 years in this dynamic, evolving industry.

Looking at Innovations in Localization

Localization thrives on innovation.

Oslo Manual defines innovation as a new or improved product or process.

Every time an LSP implements a workflow that helps clients do the translation projects better, it's an innovation:

• A client switches from emailing localization files back and forth to using a cloud-based localization management platform. Now, all files are stored in one place, translators' progress is visible in real-time, and new content is translated as the product is updated.
• A client previously struggled with manually copying and pasting translated strings from one document to another. Their language partner engineered an API to integrate the translation management system (TMS) with the content management system (CMS). Now, translated texts are sent directly to a platform that distributes them.
• Another example is automating the translation memory (TM) update process. By integrating the localization management platform with the TM system through APIs or other connectors, the system can automatically extract approved translations and update the relevant TM databases after each project. This saves time and improves consistency compared to the old manual TM update process.
• The latest trend in the language industry is the use of NLP and AI for translation. Machine translation capabilities are advancing rapidly, especially in high-resource language pairs like English to Spanish or Chinese. Some linguists now use multilingual AI to generate a first draft, which they then review and edit to produce the final text. This approach saves significant time compared to manually typing out the initial translation, creating a process innovation for linguists.
• Continuous localization as a process is innovative. It integrates translation, development, and management into a synergistic, tech-supported ecosystem where development, product management, and linguistic adaptation happen simultaneously. This approach helps evolving products remain accessible to all multilingual users at every stage of the product lifecycle.

In other words, the mindset of implementing tools to automate and improve daily operations and project delivery—making workflows better, faster, and more transparent to your team—has innovation at its heart and sets localization apart from mere translation.

The risks of too much automation: where should we draw the line?

Innovation takes many forms, and one familiar to many is process automation. The appropriate and beneficial level of automation and technology application varies across organizations and contexts. However, an "automation or bust" mentality brings several universal risks to any organization that pursues optimization for its own sake:

1. When you eliminate or reduce human oversight, the end quality might suffer. Automation without human oversight can lead to mistakes slipping through. In the world of localization, where most content is customer-facing, those kinds of mistakes can be critical.

Take a look at this screenshot from the middle of a game:

In the middle of a Polish translation, the character accuses someone of providing insufficient context—in English. Errors like this can repel players if they make it into the final product.

Another issue with reduced human oversight is the loss of creativity. Tasks requiring critical thinking and complex problem-solving are the first to suffer from over-automation. For example, in marketing localization, using AI without native-speaker oversight can lead to:<

• Style inconsistencies - shifts in tone or awkward translations that don’t align with your brand’s voice.
• Inaccurate brand tone - AI translations may be actually accurate but lack the emotion and tone of your brand’s personality.
• Dialect issues - AI varies in performance across regional language varieties and may default to the wrong version of a language, such as Brazilian Portuguese instead of Portuguese Portuguese - you won’t catch the difference unless you speak the language.
• Lack of creative translation (transcreation) - AI systems lack the shared human knowledge and cultural awareness your brand needs for accurate cross-cultural humor or reference transfer. Marketing slogans, puns and linguistic jokes are notoriously difficult to translate. This type of content requires human creativity.
• Harmful stereotypes and bias, carried over from the training data.

2. You might expose your company to security risks and potential system failures. For example, cloud automation platforms - all the online tools we’re using in work - can be the targets for cyber attacks. It’s a fantastic innovation that has changed the way we work, but is your business sensitive information secure with the platform you’re using? Evaluate the software provider’s security protocols, certifications, data privacy and compliance statements - and don’t forget to back up sensitive information stored on the platform in a reliable place.

The more tools an organization uses, the more vulnerable it becomes to system failures. Are you able to continue delivering projects if the tool you’re actively using in your workflow goes down or malfunctions? Maintain operational agility: a good strategy would be to cross-train employees to understand multiple systems and processes, not just the favorite ones.

3. When you focus on more automation, customer service and user experience may decline. Overapplied technology can save time but cause poor user experiences. Each automation decision should be assessed with the question: Will this technology help us better serve our customers? For example, automated support chatbots might provide information and you don’t have to pay them hourly wages, but complex cases often require a human specialist to actively listen, identify underlying issues, and propose in-depth solutions.

4. Automation can enter an area of ethical concerns. The most common considerations include concerns around intellectual property, data privacy, misinformation, algorithmic bias, and impact on employment.

Let’s take a closer look at one of those concerns that is particularly relevant in the language industry — the issue of algorithmic bias in machine translation.

Addressing bias in AI translation: challenges and solutions

AI translation using large language models is of those process innovations that have recently gained popularity across the localization market. This technology has transformed how organizations with strict timelines or budget constraints handle their localization needs: now they can translate large volumes quickly and at a lower cost compared to projects relying solely on human translators.

As long as these tools use public data for training, an ethical issue remains: the bias and fairness of the training content and the resulting translations.

For example, occupational stereotypes are reinforced when the gender-neutral English word "doctor" is translated into a gender-specific form in Polish, often defaulting to the masculine.

Another common issue is age stereotyping, as illustrated in this example:
 

• In an English article discussing the challenges faced by senior citizens learning new technologies, a poor machine translation might emphasize their "technological ineptitude" rather than their actual need for support and education.
   

Localization professionals working with AI translation approaches need to be well aware of the biases present in machine-generated translations.

As society evolves, so does our language about different groups of people. To further promote inclusion and support for underrepresented groups, our language must reflect these changes, especially in online content. So, what can we do to achieve this?

1. Develop with DEI in mind. During the LLM development stage, we can create algorithms that explicitly consider fairness and equity principles. Techniques like fairness constraints, adversarial training, and fairness-aware loss functions can help achieve this.
2. Provide better training data. Machine outputs depend heavily on the input data. While it may be challenging to overhaul all existing training data, we can introduce more diverse and representative content from various demographic, geographic, and cultural backgrounds.
3. Provide human oversight. To avoid age, generation, socioeconomic, gender, racial biases in the multilingual content you’re publishing, have it checked by someone who has an understanding of differences in language, and a commitment to translating content with empathy and respect. A great combination is having a machine-generation output proofread by a human translator.

Final words

We believe that human participation remains highly relevant as automation and technology enter the localization industry. The latter helps us effectively handle processes, while human judgment and creativity allow us to create high-quality, globally adaptable, and culturally relevant content that makes customers connect deeply with a multicultural brand from anywhere in the world.

While there’s no universal formula for the right amount of process innovation, it’s a journey to find the sweet spot—enough to drive organizational progress and competitiveness, but at a sustainable, balanced pace. Our approach is to implement human-centric advancements that give us more time and energy to focus on strategic and creative initiatives. How about you?