In the rapidly evolving world of AI and localization, navigating the intricacies of technology while keeping the human touch intact is paramount. That's why insights from industry experts like Morana Perić, Head of Localization at DeepL, are invaluable.
With a rich background that traverses major digital marketplaces such as GetYourGuide and Bookings.com, Morana brings a wealth of experience in enhancing user experiences and mastering the challenges of delivering multilingual content. Her journey through the mobility sector with Free Now has only deepened her understanding of the diverse needs in content localization.
In this article, we delve into Morana's expert perspectives on the fine line between neural machine translation and generative AI and their crucial roles in adapting content for global audiences. Morana's pragmatic approach highlights the necessity of marrying technology with human insight to achieve genuine localization — ensuring accuracy and cultural relevance in every piece of content.
What you will learn…
Bridging the Gap: Human Insight and AI in Localization
Artificial intelligence (AI) and machine learning have made significant advances in tasks like machine translation in recent years. However, many in the localization industry argue that more than AI is needed to ensure high-quality localized content. Humans still play a crucial role in localization that complements the capabilities of AI.
In this blog post, we explore the differences between machine translation and generative AI, their advantages and disadvantages for localization, and how humans continue to be important for localization quality assurance and earlier stages of the process.
The key takeaway is that while AI tools have become more powerful, humans remain essential partners to AI due to certain limitations in reliability and cultural understanding. Both humans and AI have complementary roles to play in localization.
Machine Translation vs Generative AI
There is sometimes confusion between machine translation and generative AI, but they differ in important ways.
Machine translation aims to match the input and output text closely by translating between two languages. Neural machine translation (NMT) models, in particular, have gotten very good at translating sentences while preserving much of the meaning and style of the source text.
Generative AI tools like GPT-3, on the other hand, provide new content based on a prompt rather than a direct translation. For example, one could prompt GPT-3 to write a blog post in German rather than inputting an English blog post for it to translate.
While both machine translation and generative AI are subsets of deep learning, they have different intended uses - machine translation for translating existing text, and generative AI for producing new text in a target language.
Advantages and Disadvantages of Machine Translation
Machine translation stays close to the source text, preserving much of the original meaning and style. However, this comes at the cost of creativity and nuance that a human translator may add. Generative AI can provide more novel and engaging outputs but is also less predictable since the content is not based on a direct translation.
Each tool also has certain technical limitations.
Machine translation works best for one-to-one translations between closely related language pairs with large parallel corpora available for training. Generative AI models today have a more limited vocabulary and understanding compared to humans. Ensuring the right tool is chosen based on the localization task is important.
“If you look at machine translation, you're closer to the source... But then you're maybe missing this, like nuance, this creativity... Generative AI... maybe gets a little bit more nuanced... But then you have this entirely new set of problems and layers of unpredictability." says Morana Perić, Head of Localization at DeepL.
Content Categorization
Not all content needs to be localized to the same standard.
Understanding the purpose and intended audience of different types of content is key to determining the appropriate workflow.
For example, help center articles aimed at answering user questions may need basic machine translation with some light editing. Marketing content intended to engage and motivate users would require more human involvement like rewriting or creative translation to localize properly.
By categorizing content, companies can optimize workflows - relying more on automated tools for lower-priority content while dedicating linguist time to content that most impacts the user experience.
This balanced approach maximizes efficiency within localization processes.
Human Role in Quality Assurance
Even with powerful AI models, humans are still needed for important quality assurance tasks.
Post-editing machine translation outputs allow humans to refine translations, fix cultural mistakes, and ensure the right tone and style.
More broadly, humans act as the final check that a localized experience will resonate properly with target audiences. Aspects like cultural appropriateness, compliance with regulations, and matching local user expectations are difficult for AI to assess without human judgment.
Partnering AI with human reviewers creates a complementary dynamic. AI handles routine tasks at scale while humans focus on tasks requiring complex reasoning, common sense, and cultural fluency. This balanced human-AI model will likely be necessary for the foreseeable future of localization.
"We are still not ready to rely on pure AI or pure neural machine translation without human intervention... You have these hallucinations that they come up with... where human intervention is invaluable," says Morana Perić, Head of Localization at DeepL.
Human Role in Earlier Stages
Interestingly, humans also play a role in the earlier stages of the localization process before AI tools are even involved. Generative AI can be used as a brainstorming tool, exploring possibilities and angles that humans may not have considered alone.
Humans then help refine and focus AI suggestions. As partners, AI and human copywriters or linguists can have thoughtful discussions to overcome creative blocks and produce the best end content. High-quality source content upstream also improves the performance of AI localization tools later in the process.
The future of localization likely involves tighter integration of humans and AI in a collaborative dynamic. AI handles routine and large-scale tasks to increase efficiency, while humans focus on work requiring human-level judgment, common sense, and creativity.
"The greatest danger of AI is that it gives very mediocre answers that are mistaken for being brilliant,” says Morana Perić, Head of Localization at DeepL.
By playing to the strengths of both humans and AI, the best localized experiences can be created to resonate with global audiences.
Navigating the AI Landscape with Expertise and Partnership
At Undertow, we expertly navigate the balance between AI and human expertise in localization to support your global expansion. With a unique mix of digital marketing savvy and localization skills, we're all about creating clear, simple solutions tailored to your needs.
We value transparency, trust, and personal connections, making every project a true partnership. Let us guide you in blending AI's efficiency with the essential human touch, creating culturally rich, authentic experiences for your audience worldwide. Choose Undertow, where technology and human insight come together for outstanding localization. Get in touch.
