I
n a rapidly evolving language technology landscape, it’s difficult for the experts to keep pace of change, never mind the laypeople. There’s never been a time of more rapid change for language technology than the artificial intelligence (AI) age, but at the same time, there’s never been a greater urgency to understand the developments as they happen.
With its Language Technology Radar, Nimdzi aims to provide exactly that logistical clarity. An interactive resource that organizes language technology into a coherent and comprehensible form, the report provides readers with the insights they need to make informed decisions — a matter of often-existential importance in pivotal years.
We spoke with Nimdzi technology researcher Yulia Akhulkova, who oversaw the report’s production, about its scope and goals. Suffice it to say we walked away with a little clarity of our own.
The Language Technology Radar Report breaks some exciting new ground for Nimdzi; let’s talk about why. Why is it considered a vital source of insights and analysis for those watching the language technology space?
Since 2018, Nimdzi has tracked the language technology space in our previous yearly report, the Nimdzi Language Technology Atlas. While the atlas served the language industry (and beyond) well by offering a unified view of the modern language technology landscape and insights into major technological advancements, it was a snapshot of the landscape of tools at a given point in time. And keeping it actualized was not exactly an easy job. When a language technology company featured in the Nimdzi Atlas changed anything, from its branding to mergers and acquisitions (M&A) activities, we had to recreate the whole “map.”
Before long, the atlas grew to over 1,000 products, and we made a decision to support it with the interactivity that this dynamic market deserves. Instead of a static technology atlas, we created a curated catalog of language technology companies. It’s no longer just a snapshot; it’s now a constantly updated database of products that brings visibility and transparency to the language technology space — and helps with related decision-making.
What was the process of building and organizing the report? How did you determine its methodology and design?
At Nimdzi, we like to write long analytic reports. So there’s nothing new here. Being privileged to have the access to insights of our incredible team of experts (which I’ve called a “supergroup” — a reference from and for my music circles), my usual challenge is deciding what to exclude from the report. All the contributions are quite insightful, but we can’t possibly include all our material in a public report: This would be a much longer report, and no one would read it!
For the design, I was inspired by portals for photographers. On websites like PhotoVogue (where I myself have a portfolio), photographers from all over the world submit their work for a chance to be featured in the curated gallery. Professional reviewers from the Vogue team accept or decline the submissions, and it’s become increasingly prestigious for an artist to be featured in PhotoVogue.
I see a similar future for the Nimdzi Radar: Month after month, more companies will submit their data. We will verify it and publish their profiles, expanding the functionalities of the online tracker itself — all to make this public resource more useful for our audience.
Who is the target audience, and how are you tailoring the information for them to maximize its value?
Technology providers are welcome to use the radar to both benchmark their competition and find partners. Investors refer to it to gain a better understanding of the leading market players. Linguists and buyers of language services can see which tools can help them in their day-to-day jobs. Students of language programs are invited to check it out to discover just how many tools may be a mere click away for use in their future careers.
This online catalog also shows that categories of language technologies overlap as they are used both as stand-alone tools and as building blocks in compound language technologies. Moreover, while some companies focus solely on developing language technology products, others bundle professional services, such as data, customization, and deployment, along with additional services. And all of them are very welcome to the radar!
You’ve organized the report into several different categories — for instance, audiovisual translation tools and translation management systems (TMSs). How did you select the categories to focus on?
We expanded the previous categorization of the atlas, which was initially influenced by the actual demand for particular tools in the language industry. One of the largest sections is TMS (see Figure 1), currently featuring over 150 solutions. Over time, we excluded some of the initial categories based on the absence of high demand. Then, in 2023, we added multilingual content generators, and in 2024, we added large language models (LLMs).
What special considerations did you take with the LLM category? What separates it from the report’s other categories?
It would be a significant oversight on our end not to dedicate a separate category on the radar to LLMs (see Figure 2), which is subdivided into four subcategories. Unlike other tools on the radar, LLMs are, by nature, general purpose. Depending on their pretraining, they can be useful in translation and localization jobs, other natural language processing (NLP) tasks such as summarization, software development, and much more.
LLMs’ general-purpose nature, rapid proliferation, and ease of access practically democratized language technologies. This resulted in a plethora of experiments with generative AI (GenAI) tools from language technology providers and tech-enabled language service providers (LSPs) to practically any tech-savvy company (or individual user). However, while buyers recognize the opportunity presented by LLMs, they don’t necessarily have the capability as such, because dealing with language and language data is different from traditional approaches.
Of which elements of the report are you the proudest?
As a researcher, I would rather run a poll to know the readers’ and users’ opinions!
I am actually quite proud that the report and the corresponding interactive tool quickly gained recognition among industry peers, and companies started submitting their data to be featured in our research. It’s an honor to see people’s appreciation, interest, and involvement in this work.
At the same time, I see much more work to be done with this project, including adding dozens of new companies and products and expanding the online tracker’s functionality and usability. I won’t disclose our big plans right now; you’ll need to wait for the release of the 2025 report with its updated online catalog!
Could you give us a bird’s-eye view of the language technology industry and where you see it heading?
Although I’ve become relatively good at reporting on what’s happening in language technology, I have never been a visionary. Could I have predicted how ChatGPT would change our lives? Not in the slightest! I did think that something like DeepSeek would make the news, but I still can’t foresee its long-term impact.
In 2023, Nimdzi created the first version of its Language AI Alphabet, which aimed to bring clarity on what is meant by LLM and GenAI, and the relationship between them (see Figure 3). With the neck-breaking evolution and adoption of core AI technologies, we soon had to update this research. The speed at which language technologies are developed is only increasing, and we expect novel categories, architectures, and tools to surface in 2025 and beyond. And we’ll have to revisit the Language AI Alphabet for updates again.
What is certain, though: Language technology solutions and products will continue to evolve. Moreover, the hype around LLM solutions triggered a new wave of investment into language technology and brought language technologies into the spotlight at the executive level. Both established companies and new language technology businesses are aiming to enhance, replace, and create language tools and workflows using AI. However, the democratization and overall accessibility of language technology, largely due to “bigtech” companies releasing similar solutions, may threaten the survival of many language AI startups.
We can also expect that, once the AI hype cools down, the “AI” prefix in company names and value propositions, now considered almost obligatory, will be dropped — just as it happened before with many other developments (remember “cloud”?).
We’ve been talking for years about the place of human labor as AI technologies continue to develop. Do you think we’re gaining clearer insights into where that balance may settle?
In my opinion, human talent in the language industry is irreplaceable. And at the same time, new tasks and roles are being created for humans. In particular, language technology providers and translation buyers may not have all the in-house expertise needed to lead AI experiments (and measure the significance of the achieved productivity and quality gains). How much and what needs to be edited? This is yet to be determined. Not to mention setting the whole experimental background, be it creating results-oriented workflows, evaluation metrics, and benchmarks; checking the appropriateness of LLM outputs for particular use cases and content types; training AI agents; and much, much more.
A subsidiary factor in this conversation is quality. How can translation quality be assessed, and how does language technology influence the process of achieving the desired quality?
With AI of choice at hand, businesses now have a complex task to identify whether there are significant gains in productivity for their particular use cases and verticals. And productivity in this case is a function of quality. When assessing AI translation quality in the LSP world, people often talk about the same quality metrics that apply when evaluating human translation services (accuracy, fluency, style, consistency, domain-specific terminology, turnaround time, etc.).
In the language technology world, metrics like BLEU (Bilingual Evaluation Understudy), METEOR (Metric for Evaluation of Translation with Explicit ORdering), TER (Translation Edit Rate), chrF (Character n-gram F-score), and hLEPOR (hybrid Language Evaluation of Precision, Order, and Recall) and custom in-house metrics are used to evaluate machine translation (MT) quality, also helping to adjust the training of MT engines to improve over time. For LLMs, metrics choice depends on the goal. For example, BLEU or chrF can be applied for quick checks; BERTScore (a computationally heavier metric using contextual embeddings from models like BERT, which compares semantic similarity between translation and the reference at a “deeper” level) or COMET (Crosslingual Optimized Metric for Evaluation of Translation) for meaning; and certainly, human evaluation for everything. When evaluating LLM output, researchers also use pass/fail tests.
All in all, automatic metrics are useful for quicker and large-scale evaluations, human evaluation is used for assessing translations in-depth, and then there are newer methods that attempt to combine the best of both worlds, leveraging pretrained models for a deeper semantic understanding.
Incorporating human intervention is crucial not only during stages like editing or final review for balancing speed with quality but also when developing clear metrics and benchmarks of AI-powered translation, as well as estimating its actual return on investment.
What do you see as the greatest opportunities for language technology and the broader industry in this tumultuous time? And what are the greatest challenges?
One of the most significant challenges language technology providers face today is effectively selling their sophisticated technology products. The democratization of language technology has led to a market saturated with options, making it increasingly difficult for providers to differentiate themselves and convince potential buyers of their product’s unique value proposition.
Language technology companies must adopt innovative approaches not only in their development but also in their targeted marketing campaigns, which should highlight specific use cases and benefits, as well as in building strategic partnerships with complementary technology providers to offer bundled solutions. We see this strategy both in the integrators and connectors space and in the sector’s reactivated M&A.
Ultimately, success in the market will depend on a provider’s ability to adapt to the changing landscape and develop innovative sales and marketing strategies that resonate with potential buyers. And it’s not easy, if you ask me — which is proven by the fact that, at Nimdzi, we see more and more language technology providers requesting consultancy and help in expansion and growth.
Is there anything else you want to add?
I’d like to use this opportunity to express my gratitude to the Nimdzi team, particularly Renato Beninatto and Josef Kubovsky for inspiring me to do what I want at work (how lucky am I?), Laszlo Varga for his professional review and the magic goblet of insights, and most importantly, Aleksey Schipack, who made the online tracker possible. Aleksey usually stays behind the curtain of Nimdzi work, but without him, there wouldn’t be the Nimdzi Language Technology Radar as we know it.
If your company’s language technology solution is not yet featured on the radar, make sure to submit your data via https://www.nimdzi.com/add-ltr/.
Related Articles
Since 2018, Nimdzi’s annual flagship report has been accessed more than 65,000 times, underscoring its role as a cornerstone of industry intelligence with a commitment…
→ Continue ReadingThis week’s stories underscore a shared commitment to building multilingual futures that last, where the industry is constructing frameworks that will define global communication for…
→ Continue ReadingThe Nimdzi 100 Ranking of the most successful translation and localization companies globally has investors looking towards the localization industry.
→ Continue Reading
