The rapid development of AI in localization
AI and machine translation have made giant leaps in recent years with tools like Google Translate, DeepL, and ChatGPT, which have revolutionized the way we approach translation and thus localization. But is it enough to guarantee an optimal user experience and perfect adaptation to local markets? You know concretely that no, translations are not contextualized enough and, in some cases, poorly translated. You can also find several examples on our blog in the article
AI and machine translation have made significant advancements in recent years with tools like Google Translate, DeepL, and ChatGPT, which have revolutionized the way we approach translation and localization. However, is it sufficient to ensure an optimal user experience and perfect localization?
You can find several examples on our blog in the article “*5 Critical Localization Mistakes That Could Tank Your Global Success”*.
Why integrate AI in your localization strategy?
We have observed three major benefits for our users by integrating AI into Gleef:
- Time Savings and Cost Reduction: AI handles most of the translation work, with human verification ensuring quality.
- Improved Consistency: Modern AI models standardize tone and vocabulary across all translations.
- Customization for Local Markets: AI can now adapt content based on cultural context, ensuring better engagement.
These benefits align perfectly with the strengths of our plugin, which allows you to translate your keys effortlessly:
- AI-generated automation to save time: Automate translation, key generation, and retrieval.
- Workflow optimization: Ensure clarity and consistency while avoiding duplicate keys.
- Project organization: Track translated and pending keys to manage localization projects efficiently.
- Code quality: Maintain uniform formatting of localization files for a professional output.
Challenges to Overcome
While AI brings speed and scalability to translation, several limitations remain:
- Semantic ambiguity and context errors: AI can misinterpret words with multiple meanings if context is unclear. For instance, the English word "remote" was mistranslated into French as "télécommande" (TV remote) instead of the more appropriate "télétravail" (remote work), leading to confusing or misleading content on a job advertising platform.
- False friends and lexical traps: AI systems can be tripped up by false cognates — words that appear similar across languages but have different meanings. A common example is "save", which was mistakenly rendered as "économiser" (to save money) rather than "enregistrer" (to save a file) on the Uber app.
- Machine translation and workflow integration: While AI speeds up the initial draft, translations often require human post-editing to ensure accuracy and natural tone. In practice, this can lead to back-and-forth revisions, and integrating AI into existing translation workflows presents technical hurdles — from software compatibility to training and data privacy concerns.
A real-world example: A translation company mistakenly relied on AI to translate its log-in page from English to French. The result? Instead of the natural "Ravi de vous revoir" for "Welcome back," the system produced the awkward "Bon retour," a literal translation of "Good return" that feels unnatural in context. This underscores how AI-driven translation can misinterpret nuances, leading to clumsy phrasing that disrupts the user experience.
How to successfully integrate AI
At Gleef, we leverage AI to automate localization while ensuring high translation quality. Traditionally, localization workflows generate keys in two ways: using element names or unique identifiers. However, both methods have significant drawbacks.
The first approach relies on element names or their textual content, which quickly results in unreadable names and duplicates when elements are reused. The second involves generating unique identifiers that avoid naming conflicts but offer no insight into the content, making management and reuse more complex.
To address this, we implemented Retrieval-Augmented Generation (RAG), which combines contextual search with AI-powered key generation. This approach provides structured, understandable, and reusable keys tailored to each project’s needs.
Our system relies on three essential principles:
- Reuse of existing keys: Prevents duplicates and ensures consistency.
- Intelligent, contextual generation: AI follows adapted naming conventions.
- Consistency and organization: Enforces homogeneous nomenclature and facilitates maintenance.
Gleef customers can also integrate their existing key databases, enabling us to leverage current resources rather than generating new entries from scratch. This enhances consistency and eliminates unnecessary redundancy. To explore our product in detail, check out our article: “Keys, Translation, and Context: Dive Into the Heart of Our Plugin”.
Once configured, our plugin scans Figma elements to identify each component — buttons, placeholders, labels, and even screen structures — ensuring keys are precisely adapted to their context. The result? A seamless workflow, well-organized keys, and effortless management—all powered by AI.
AI as a time-saving tool
AI-powered translation significantly reduces time spent on UX writing by minimizing back-and-forth. Based on our research and experience, a manual translation with validation typically takes 30 seconds to draft the wording, 1–2 minutes to send and wait for a response, and another 1–3 minutes for adjustments—totaling 3 to 5 minutes per message. In contrast, using an LLM takes just 10 seconds to request a translation, 20 seconds to review it, and 10 seconds to implement, reducing the total time to around 40 seconds.
When fully automated with Gleef, the entire process can take as little as 5 to 10 seconds, representing up to 90% time savings per wording. Given the volume of wording per feature, the cumulative time savings are even more significant. Integrating AI-powered translation directly into designer workflows is a highly effective way to eliminate delays and streamline the overall process.
Conclusion
AI is a game-changer in localization, but its effectiveness depends on proper implementation. With Gleef, we aim to simplify multilingual content management while maintaining human oversight for quality assurance. We are continuously optimizing our approach and value community feedback.
What are your localization challenges? What tools do you use? Share your experiences in the comments, and let’s shape the future of AI-powered localization together!
