For the past two years, the same question has been running through the entire translation industry: what is left for humans when artificial intelligence can translate in a matter of seconds? Many see it as a threat. Others have seen it as an invitation to rethink everything — and the results are promising.
Here is what is changing, in practical terms, both for those who commission translations and for those who produce them.
The problem everyone knows
Translating a document is not simply about converting words from one language into another. It means finding the right term — the one the client approved three years ago, in another project, on another platform. It means ensuring that a user manual, a software interface and a training video all speak exactly the same language. It means never doing the same work twice.
Traditional tools such as Trados, memoQ and Déjà Vu provide partial solutions, but at the cost of cumbersome proprietary environments and expensive licences.
The emerging generation of platforms no longer searches simply for similarities between words: it searches by meaning. Traditional translation memories primarily look for similarities between formulations. Next-generation memories add another dimension: they look for semantic similarity.
A sentence that has been rephrased, reordered or expressed using different terminology can therefore retrieve a relevant translation that text-based similarity alone might have missed.
Consider these two formulations, for example:
Stop the engine immediately.
The engine must be shut down without delay.
A traditional translation memory may consider these two segments significantly different because they share relatively little textual similarity. A semantic translation memory, by contrast, identifies their similarity in meaning and can retrieve the corresponding validated translation.
And when no validated translation exists yet, these platforms can use artificial intelligence to generate a first version based on the actual context, leaving the linguist to refine it.
For companies: faster, more accurate and increasingly cost-effective
If you commission translations, three things change.
Turnaround times become shorter because the entire production process is streamlined, and a large urgent project can be distributed among several linguists working simultaneously without sacrificing consistency.
Costs decrease progressively: every translated project enriches your own translation memory and increases the potential for reuse in future projects. Your translation history becomes an asset that works for you.
And finally, quality ceases to be merely a promise and becomes something that can be measured. Quality control is based on the international MQM (Multidimensional Quality Metrics) framework: quality is no longer simply claimed; it is assessed, objectively measured and traceable. This is a decisive advantage in sectors where translation errors can have serious consequences.
For agencies: freedom without the constraints
If you are a language service provider, next-generation translation memories free you from proprietary environments. There is no longer any need to impose a proprietary environment or locally installed software licence on every user.
Your existing translation memories, whatever their origin, can be recovered and put back to work. They are automatically enriched throughout the translation and post-editing process, with no manual importing required, and are automatically organised by client.
You can monitor the progress of every project in real time and, with your clients' consent, take advantage of the combined wealth of your translation memories.
For linguists: all the convenience, none of the constraints
And because a platform is only as good as the people who use it, the translator's day-to-day work also needs to be rethought.
Everything can now be done through a simple web browser, with no licence to install, while retaining the full range of features expected from a professional translation environment: contextual search, glossaries, segment filtering and progress tracking.
The power of an expert tool, with the simplicity of a web page.
Innovation without dependence on proprietary cloud platforms
Major players in the translation industry are already beginning to offer this new generation of translation memory through proprietary cloud platforms. But this increases users' dependence not only on proprietary tools, as in the past, but also on the infrastructure operated by those providers.
Previously, translation environments could continue to run locally as long as the user held a valid licence. In the future, users may need not only a licence but also continued access to the provider's proprietary infrastructure. Dependence therefore extends beyond the software itself to the infrastructure on which it runs.
But technological progress now allows smaller players to compete with the largest. Open-source resources give them the means to develop such platforms by drawing on the immense ecosystem of open-source technology.
All they need is access to the most widely distributed resource in the world: human intelligence.
Tradoc has chosen this path and now operates a platform offering the capabilities described above in a local environment, on its own servers, ensuring both its independence and the confidentiality its clients expect.