REVIEW

The translation management systems (TMS) market stands out prominently within the language services industry as one of the most bespoke technology spaces. With its rich history, TMS ranks among the oldest language technologies, witnessing the emergence of STAR Transit and Trados in the 1980s and the systematic transformation of CAT tools into extensive translation management platforms that allow both buy-side and supply-side localization stakeholders to manage the translation process of multiple content streams effectively. In 2020, the acquisition of SDL by RWS set the tone for a strong M&A and investment wave that would last until mid-2021 and affect many companies, including Across, Lingotek, Memsource, and XTM. Today, TMS remains the cornerstone of translation technology, serving as the glue that holds together various components such as translation memory, terminology, machine translation, quality assurance, and different project management assets.

While the appetite for acquisitions and venture capital has cooled, the market continues to experience substantial growth, resulting in an estimated market size of $0.3 billion. Even though the competitive landscape is dominated by the industry’s super agencies — TransPerfect (GlobalLink) and RWS (the Trados ecosystem) — it is more diversified than ever. Established players offering generic TMS like memoQ, Phrase, and XTM Cloud are steadily growing and increasing their footprint. TMS focused on continuous localization of software strings are gaining market share, confirming that developers have become prominent translation buyers. An array of new niche products has emerged, including Weglot, a user-friendly website localization solution for marketeers and website owners, and Cleverso by Tarjama, catering to the MENA region. Finally, there are TMS with LSP grassroots like Bureau Works and open-source systems such as translate5.

The race for generative AI adoption

TMS companies rushed to add LLM-based features to their product portfolio. Crowdin, Lokalise, and Trados integrated AI assistants; Bureau Works, memoQ, and Smartling announced LLM-driven machine translation engines, whereas LILT released a self-service model-building and training studio. One of the most observable trends is that pure translation technology providers such as Smartcat, DeepL, Writer (formerly Qordoba), and LILT have embarked on introducing AI copywriting tools, a strong endorsement of leveraging LLMs for multilingual content creation.

Despite the plethora of initiatives, we haven’t determined a significant business impact yet. Furthermore, the top TMS players are cautious as almost every AI-powered feature released across the different players is based on OpenAI, which is relatively easy to replicate and increases the risk of rapid obsolescence. For now, buy-side localization stakeholders are teased with demos, trials, and webinars. For tomorrow, the industry expects integrations with different models and impactful use cases that positively affect how translation memories, terminology, quality assurance, and automated translation are leveraged.

The revival of the translation memory

The big neural machine translation push of 2017 enticed MT advocates to primarily look at translation memories as data tanks used to train NMT models. Despite the systematic adoption of machine translation across many corporate localization departments and LSPs, MT has not succeeded in displacing translation memories from their position as the primary linguistic resource in translation workflows. The advent of LLMs could imply that translation memories are once again in danger of being relegated to the background.

Surprisingly or not, we’re observing the opposite effect. The rise of GenAI has spurred general interest in using LLMs as the broom to clean old translation memories. While translation memories are often exposed to linguistic pollution, especially when used over a long period of time, LLMs can identify obsolete and low-quality units following a set of company-specific, domain-specific, or product-specific instructions, eliminating the need to spend hours of manual work on tedious cleanup operations. Another use case where TMs and LLMs are mashed is fuzzy match augmentation — a noteworthy one since fuzzy matches are essentially translations with a fixed error rate.

Paradoxically, no matter the use case, the ultimate goal is to continue using translation memories as the main translation helper in localization projects. The greater the impact of AI, the stronger our interest in traditional linguistic resources becomes.

Orchestration: the new buzzword

The TMS connector, an important revenue-driving asset in the product portfolio of many TMS providers, is under siege as it is criticized for its limited point-to-point use cases and lack of customization. The new buzzword in the realm of automation is orchestration, whereas the focus is shifting towards end-to-end integrations of the different applications in the localization tech stack. Third-party middleware providers like BeLazy and Blackbird.io have recognized this need and provided refreshing alternatives to the bulkier products in TMS connector pools. Phrase follows suit with the launch of Phrase Orchestrator.

While our industry has seen many attempts to solve the connectivity problem for the non-technical localization professional, it remains a very hard nut to crack. With the new year and the increased competition in the connector space on the horizon, the challenge can be addressed with a fresh perspective.

Data-driven localization

Given the current economic climate, data-driven localization is becoming imperative to measure ROI, monitor KPIs, and make informed strategy decisions. Phrase and memoQ have released customized reporting features with Analytics and Business Analytics, respectively, thereby reviving earlier initiatives observed in other TMS such as GlobalLink and Lingotek. We anticipate that other TMS providers will follow the trend and announce similar reporting functionalities.

At the same time, 2023 has been the year of the breakthrough of machine translation quality estimation (MTQE). MTQE is intended to predict how much post-editing a specific machine translation requires, thus identifying potential risks associated with the translation. Apart from being cost-effective and fast, it is also scalable since the data used to train MT models can be repurposed to train MTQE models. It’s a great opportunity for localization managers with lots of MT-ready content, such as ecommerce product owners. Leading providers include Unbabel, TAUS, and ModelFront. In the TMS space, Smartling pioneered forward-looking quality scores in 2017 through quality confidence scores (QCS), but had to pass the torch when Phrase’s proprietary MTQE mechanism went live. memoQ facilitates MTQE by integrating ModelFront and TAUS.

Failure to adapt to multimedia localization needs

Many TMS providers, from memoQ and Trados to Lingotek, Smartcat, and XTM, ventured into subtitling by enabling live video previews in their workbenches. Meanwhile, the interest in AI dubbing is increasing at an incredible pace for several obvious reasons: It’s fast, customizable, relatively inexpensive, baseline quality is excellent, and there are good use cases for it. Think about elearning. We saw a similar discipline, machine translation, gaining a critical mass for the exact same reasons.

While we can no longer count the number of tools with dub and .ai in the name (deepdub.ai, dubdub.ai, dubformer.ai, dubverse.ai — the list goes on), the top TMS players are seemingly losing their grip on audiovisual localization in favor of specialized localization suites and AI-powered dubbing services. Given the demand for synthetic voices is skyrocketing, the first TMS provider to move into text-to-speech (and, optionally, speech-to-text) will have a considerable head start.

2024 outlook

The year 2024 promises key GenAI and LLM breakthroughs in the TMS market. It is undeniable that the introduction of LLMs allows the language services industry to rethink and perhaps redesign some of its most traditional disciplines. Think about QA; running QA checks is a time-consuming practice that generates many more false positives than real errors, while the underlying rule-based algorithm cannot learn from ignored false positives or transfer them from one project to another. It’s a perfect fit for LLMs.

Nonetheless, 12 months after the big introduction of ChatGPT, many localization stakeholders are still in the exploratory phase. It’s not unlikely that many people will ever get past this phase or even operate an LLM outside ChatGPT. Ultimately, the localization industry is a community of language professionals, and translation is its core activity. Research and development are reserved for tech companies and academics.

With the market’s sustained growth and diversification, TMS is firmly established in the heart of language technology and a significant part of the industry. Buyers as well as service providers will wait for the TMS companies to facilitate access to AI-driven applications, integrate models from providers beyond OpenAI, and develop use cases. If all the weight is put on the shoulders of the TMS market, GenAI and LLMs will become nothing but new tools for existing processes, like translation and localization.

BEST SUITED FOR:

• Enterprises needing advanced localization workflows and reporting in a server scenario

• Technical writing teams using document CMS

• LSPs with security and automation requirements

DEPLOYMENT MODEL(S):

• Private Cloud

• Public Cloud

• Server/On-premise

API:

REST
SOAP

BEST SUITED FOR:

• Large ecommerce projects with clever batch distribution, batch progress and auto-mated tracking of repetition

• Arabic-centric projects

DEPLOYMENT MODEL(S):

• Private Cloud

• Public Cloud

• Server/On-premise

API:

REST

BEST SUITED FOR:

• Companies and organizations that frequently improve their multilingual products

• Localization managers, developers, and other teams working with content can manage and automate the localization process

DEPLOYMENT MODEL(S):

• Private Cloud

API:

REST

BEST SUITED FOR:

• Enterprise localization teams with volumes in millions of words and advanced multi-step workflows

DEPLOYMENT MODEL(S): 

• Private Cloud

• Public Cloud

• Server/On-premise

API:

REST

BEST SUITED FOR:

• Mobile app developers

• Growth-oriented companies with tech-savvy teams

DEPLOYMENT MODEL(S):

• Public Cloud

API:

REST

BEST SUITED FOR:

• Organizations interested in high translator satisfaction and productivity

• LSPs that receive localization packages from multiple buyers in different formats and want to process them in a single interoperable system

• Buyers with proprietary systems that require custom integrations

DEPLOYMENT MODEL(S)

• Private Cloud

• Public Cloud

• Server/On-premise

API:

REST

SOAP

BEST SUITED FOR:

• Financial companies

• Legal localization

• Global marketing Public Cloud

DEPLOYMENT MODEL(S)

• Public Cloud

• Server/On-premise

API:
REST

BEST SUITED FOR:

• LSPs

• Translation customers looking for linguists/LSPs to streamline processes

• Freelance and in-house linguists, as well as related professionals

DEPLOYMENT MODEL(S):

• Private Cloud

• Public Cloud

API:
REST

BEST SUITED FOR

• Enterprise and midsize companies with new or established localization programs, seeking automation, quality, and cost-optimization in their translation process.

• Companies seeking expert translation services, machine translation, and AI-powered translation options.

DEPLOYMENT MODEL(S):

• Private Cloud

• Public Cloud

API:
REST

BEST SUITED FOR:

• Translators

• Project managers

• Companies looking for a TMS that offers:

• Cloud-based TMS

• Agile processes

• Customizable interfaces

DEPLOYMENT MODEL:

Public Cloud

API:

REST

BEST SUITED FOR:

• LSPs

• Industrial companies

• State/government institutions

DEPLOYMENT MODEL:

• Private Cloud

• Public Cloud

• Server/On-premise

API:

REST

BEST SUITED FOR:

• Corporate translation and localization departments with complex needs in automation and reporting

• Medium to large LSPs that want to keep their translation activity in a single TMS and CAT tool

DEPLOYMENT MODEL

• Private Cloud

• Public Cloud

• Server/On-premise

 API:

REST
SOAP