Welcome to the LOCANUCU blog, your home for localization news you can actually use. In our roundup for July 11, 2025, we're covering major developments across the industry. We'll look at a significant shift in how the Canadian government is buying translation services, peek behind the curtain at the incredible computing power DeepL uses to train its AI, and explore China's new focus on "linguistic self-reliance." We also have updates on specialized AI for fiction, new professional training programs, and critical advice on avoiding common localization pitfalls. Let's get into the details.
- Canada's Translation Bureau is piloting a new "Agile Bidding" procurement method.
- The new Canadian system will use a pool of pre-qualified suppliers instead of per-project bidding.
- The goal of Canada's procurement shift is to improve translation quality and supplier relationships.
- China is reducing the emphasis on foreign language education in schools to promote "linguistic self-reliance."
- The European Language Industry Association (Elia) has released a new report on the EU's translation and interpreting business.
- A new service, WONDER AI, has launched to provide specialized AI translation for fiction, novels, and comics.
- DeepL revealed it uses NVIDIA DGX SuperPOD infrastructure to train its large-scale AI translation models.
- Progress is being made in integrating Uzbek machine translation (MT) into CAT tools for translators.
- A SlatorPod episode has spotlighted the top 50 language AI startups to watch in 2025.
- Stefan Huyghe argues the biggest flaw of LLMs is their lack of a world model and true understanding, not hallucinations.
- LangOps platform Translastars has introduced a new training program for localization professionals.
- The tutors for a new Master's in AI Innovation for the Language Industry have been announced by Dorota Pawlak.
- A Lokalise guide emphasizes the importance of a structured process for finding and vetting localization vendors.
- Effective vendor management includes building strong, long-term partnerships, not just focusing on cost.
- Localazy identifies hard-coded strings as a major technical obstacle in software localization for fast-growing companies.
- Another common software localization pitfall is the lack of a centralized platform, leading to chaotic workflows.
- Josef Kubovsky shared an example of a personalization fail where an email included the placeholder `[COMPANY_NAME]`.
- Helen E. Song discusses the "Silent Partner Protocol," analyzing how check interpreters can disrupt interpreting sessions.
- A study from Preply shows many Americans are confused by common online acronyms like 'SMH' and 'OOMF'.
- Routledge has published "On Translation," a new book on translation studies by author Xu Jun.
In government and procurement news, Canada's Translation Bureau is taking a new approach to acquiring translation services. The bureau is launching a pilot program to move away from traditional competitive bidding for individual projects. Instead, it will establish a pool of pre-qualified suppliers. This "Agile Bidding" method aims to improve the quality and consistency of translations by fostering longer-term relationships with vendors, a significant shift designed to enhance efficiency and partnership within the public sector's language services.
Staying on the topic of governmental language strategy, a notable trend is being observed in China, where a push for "linguistic self-reliance" is leading to a decreased emphasis on foreign language education, particularly English, within the national curriculum. This policy shift could have long-term consequences for the global talent pipeline and the localization industry's access to bilingual professionals in the region.
On the European front, the European Language Industry Association, Elia, has released a new comprehensive report. The study, titled "The Business of Translation and Interpreting in the EU," offers fresh data and insights into the market landscape, providing valuable benchmarks for language service companies operating within the European Union.
Shifting to technology, DeepL has offered a glimpse into the immense power required to train its advanced AI translation models. The company detailed its process for deploying NVIDIA DGX SuperPOD infrastructure, highlighting the significant investment in high-performance computing necessary to stay at the forefront of machine translation quality. This underscores the resource-intensive nature of developing top-tier AI language solutions.
In the realm of AI applications, a new service called WONDER AI has been launched, specifically targeting the translation of fiction, including novels, comics, and webtoons. This specialized service aims to tackle the nuanced and creative challenges of literary translation that general-purpose MT systems often struggle with. For less-resourced languages, there's positive news regarding Uzbek. Recent developments have focused on improving and integrating Uzbek machine translation directly into CAT tools, a crucial step for increasing efficiency and support for the language's translators.
Looking at the broader AI startup landscape, a recent SlatorPod episode identified the 50 top language AI startups to watch in 2025, signaling vibrant innovation across the sector, from AI dubbing to advanced speech technologies. However, not everyone is focused solely on capabilities. Industry expert Stefan Huyghe shared a critical perspective on the limitations of Large Language Models, arguing that their biggest flaw is not hallucination, but rather their fundamental lack of a world model and genuine understanding, a crucial distinction for those implementing AI solutions.
In professional development, Translastars has launched a new training program through its LangOps platform. The program is designed to equip localization professionals with the evolving skills needed in a tech-driven industry. Furthering the integration of AI and education, Dorota Pawlak announced the roster of tutors for the new Master's program in AI Innovation for the Language Industry, indicating a growing academic commitment to formalizing these critical skills.
Turning to professional practice and common challenges, a blog post from Lokalise offers a guide to effective vendor management in localization, covering the essentials from vetting partners to building strong, lasting relationships. In a similar vein, Localazy has outlined six common pitfalls that rapidly growing companies face in software localization. These include issues like hard-coded strings and the absence of a centralized platform, serving as a warning for scaling businesses. On a more granular level, a post by Josef Kubovsky from Localazy highlighted a simple but damaging mistake: a company sending a marketing email with the placeholder [COMPANY_NAME] still visible, a clear failure of personalization and localization QA.
In the interpreting field, Helen E. Song introduced a discussion around the "Silent Partner Protocol." This protocol addresses how a check interpreter's interjections, while well-intentioned, can sometimes derail the primary interpreter's flow and disrupt a session, sparking a conversation on best practices for collaborative interpreting assignments.
From a linguistic and cultural perspective, a recent study by Preply revealed that a significant number of Americans are confused by common text and online acronyms. This highlights the rapid evolution of digital language and the communication gaps that can emerge, a relevant reminder for anyone working in localization and communication.
Finally, for those interested in the academic foundations of our field, publisher Routledge has released a new book by Xu Jun, simply titled "On Translation." The work promises to be a significant new contribution to translation studies, exploring the theoretical underpinnings of the craft.
That's a wrap on today's key developments. The major takeaways are the industry-wide trend towards more strategic partnerships, as seen in Canada's new procurement model; the astronomical computing resources now required to compete at the highest level of machine translation; and the persistent need for robust processes to avoid embarrassing and costly localization errors. The conversation around AI's true "understanding" and the evolution of digital language continues to shape our work. For more actionable insights and to stay ahead of the curve, keep following us at LOCANUCU - Localization news you can use.