In the period between May and July 2024, publicly listed game services and localization provider Keywords Studios has been in the news about possibly going private. It became known in May that Swedish private equity firm EQT had made a GBP 2.2bn (USD 2.8bn) offer for the company. And as of the publication of this article, the companies had reached an agreement for GBP 2.1bn (USD 2.7bn).

The acquisition of the London-listed Keywords Studios was recommended by the Board, and it is subject to shareholder approval. If it closes, the deal would give EQT 51% control of the company. The rest would be evenly split among the two other stakeholders, co-investors CPP Investments and Rosa Investments.

There is considerable interest from private investors in language technology and language services (or a combination of both) as the first half of 2024 comes to an end, with firms like Valorem Group and Mayfair Equity Partners as examples of major fund infusions.

We asked readers if they thought we would see a major language industry IPO this year, and while there are no definitive signs that this can happen, most (66.7%) said Yes. The rest (33.3%) said No.

Is ChatGPT Now an Everyday Tool?

Maybe. For about a third of Slator readers, anyway. 

With OpenAI staying very present in people’s minds, through actual AI technology news or some new drama unfolding on X, it is easy to forget that ChatGPT will celebrate the second anniversary of its public launch this year. The question asked online about the LLM has gone from “Have you tried it?” to “Have you tried it for this or that?” to “How are you using it?” 

That includes Slator’s Reader Polls, answered weekly in our language industry newsletter, with ChatGPT having de facto relevance because, well, the bot can translate. We first asked readers in December 2022 how big of a deal ChatGPT was, and over a third (37.9%) said back then it was a “very big deal.”

We have continued asking readers periodically about their use of ChatGPT, including whether that use included machine translation (MT), ChatGPT’s impact on the language industry in the short term, and more. As of July 2024 the model has significantly evolved and the latest Slator question to readers about it has evolved as well: How has your use of ChatGPT / other LLMs developed so far in 2024?

For a little over a third (33.3%) of readers, the use of ChatGPT stayed the same. Use increased a lot for over a quarter (28.9%) and a bit for under a quarter (24.5%). There is an even split among readers whose use decreased a bit (2.2%) or a lot (2.2%). The rest (8.9%) stopped using it altogether.

TMSs at a Crossroads

The production side of language services has heavily relied on the tried and true features of translation management systems (TMSs) since the 1990s. And until neural machine translation entered the localization process, the general structure of TMSs underwent little change. 

Things are very different in July 2024. Machine translation (MT), now enabled by AI, is but a small component of the translation and localization cycle, and the management aspects of the process can all now be highly automated and integrated using AI. 

While a few of the well-established TMSs have incorporated some level of automation, new products continue to enter the market, at the same time driving localization buyer expectations. A look at AI orchestration for localization, for example, can alone serve as an example of what is now possible.

We asked readers if they are happy with their TMS, and most responders (48.0%) said “not really, needs improvement.” Over a third (36.0%) believe their current choice does the job, and the rest are content (16.0%) with it.

Tell My Agent to Do It

A look at the industrial robots working car assembly lines (like entire sections of the Volkswagen mega factory in Wolfsburg, Germany), or at those doing visual quality assurance (like the ones made by COGNEX and Mitsubishi), implies that having their virtual AI counterparts work a fully automated cognitive stream of tasks is no novelty.

In fact, Andrew Ng, a recognized AI pioneer and expert researcher, created a prototype for AI agentic translation, released on June 11, 2024, as an open-source demo. In this case, the tasks are within a machine translation (MT) flow, and an agent asks an LLM (e.g., GPT-4 turbo) to translate text and then “reflect” on the output. This is not too different from functionalities already integrated into some of the newer AI translation systems, which can either self-correct or learn from human MTPE as a form of supervised machine learning. 

Since all robots, physical and virtual, are built for task completion on the basis of predictability, the question is not whether a fully working automated process learned and run by AI agents is possible. That is a given and Ng showed it. The real question is, when and how would it be truly operational for commercial localization use cases? For what types of content and projects? And how can it be scaled?

Those and other thoughts are in the minds of many, but surprisingly, most Slator readers (43.8%) are actually not that interested in agentic machine translation (“Meh”). A little over a third find the notion interesting (37.5%) and the rest (18.7%) think it is the next big thing.