January 2020
Text by Sara Grizzo

Image: © cybrain/istockphoto.com

Sara Grizzo is a freelance translator and post-editor based in Munich, Germany. She offers training in machine translation post-editing for companies and fellow translators and loves to share her knowledge and experience in conference presentations, magazine articles, and her LinkedIn series.




The challenge of being a translator – and a client – in the age of AI

The arrival of Neural Machine Translation has turned the translation industry on its head. Time for companies to reap the opportunities; time for translators to find their place in this new landscape.

Before Artificial Intelligence (AI) and Neural Machine Translation (NMT) materialized in the localization and translation world, the landscape of our industry was pretty straightforward. When it came to translating content written in language A into language B, two scenarios would open up:

In the first scenario, clients required spot-on translations that needed to be perfect in terms of terminology consistency, style adequacy, punctuation, grammar and spelling. In this case clients cooperated with professional linguists who relied on the different functions of CAT tools (translation memories, glossaries and QA options) in order to provide the necessary quality.

In the second scenario, clients chose to use machine translation (MT) – rule-based at first, followed by statistical engines in the early 2010s – for handling translation assignments that either involved highly standardized content, i.e. with relatively simply built sentences and no room for ambiguity, or required only a so-called "good enough" quality; this was often the case for gisting and internal communication. Of course, MT output needed to be checked and improved, so clients depended on translators willing to take care of the post-editing (PE).

Back then, translators weren’t very keen on this kind of task; some of them reacted quite emotionally when the topic was brought up. Only a relatively small group of linguists were involved in post-editing projects. Many would turn down such assignments, mainly because they feared the work on the output would take way too long and not bring in enough money.

Then came the true turning point for our industry, one that changed the landscape forever. Between the end of 2016 and the beginning of 2017, Google and DeepL announced they had eventually succeeded in developing their neural engines.

The arrival of the Babel fish

These new engines have been able to perform much better than previous systems. They have delivered an undeniable improvement in terms of output fluency, a better understanding of the source text as well as the ability to make out mistakes in the source text. In addition, NMT can cover a larger array of language combinations. Despite the downsides of this approach – lack of terminology consistency, added/missing content and plausible but misleading translations – the latest technology bears great potential.

For clients, this new development means they can apply MT to complex content, to more content types, to "new" language combinations as well as to different domains and applications. As a result, we now have plenty of different scenarios and a more multifaceted understanding of quality that very much depends on the client requirements, project specifications and/or turnaround times:


  • Human translation, with linguists translating the source text in the traditional way, only relying on the technology aid provided by CAT tools
  • Post-editing within CAT tools, where the bilingual files have been pre-populated with MT output (in this case quality requirements can range from light PE to full PE, with clients often preparing very clear guidelines about the standards the final text needs to fulfill and the mistakes that can be ignored)
  • Post-editing within online tools, often developed by the clients themselves and integrated into the company workflows
  • Human translation supported by machine translation and predictive typing solutions in order to improve the linguists’ productivity (so-called "augmented translation")


But what about the translators?

Certainly, translators today can’t deny the recent advances in technology and the improved output quality of the NMT engines, at least in some domains. With plenty of MT solutions available online and as plug-ins for almost every CAT tool, translators can run ample tests on different content and draw their own conclusions. People are no longer laughing at the online translations suggested by Google or other providers. Overall, linguists have developed an increasingly positive attitude towards MT. Many have already implemented it to maximize their own productivity. Or to get some inspiration in case they aren’t 100 percent sure about the meaning of a specific sentence or just need some brainstorming assistance around a certain term. For instance, the Thesaurus function available on the DeepL online interface is very popular among linguists.

But despite the obviously good performance of NMT and its potential as a productivity boosting tool, the translator community is generally unsettled by the changes occurring in our industry. Specifically, translators


  • fear that at a later stage they will be replaced by the machines and will eventually become redundant;
  • aren’t pleased about machine translation being suddenly applied to every kind of content, especially those less suitable for MT, such as marketing content or creative texts;
  • are aware of the hidden risks of NMT (added/missing content, output that sounds plausible but doesn’t reflect the source text) and reluctant to take on post-editing because they are afraid of overseeing major errors and of having to take liability for these kinds of mistakes;
  • find it generally difficult to take on light PE assignments as they are not used to delivering lower quality and to ignoring basic grammatical mistakes or style issues;
  • don’t like the idea of putting up with the MT output, providing them neither freedom to build their own sentences nor room for creativity;
  • have concerns about the long-term impact of MT usage on the variety and liveliness of the target languages;
  • feel a huge pressure on their rates and more generally on their productivity, because in general the turnaround times of translation projects are shortening drastically.



Starting the dialogue

Sure enough, NMT is a rather disrupting development for our industry that offers great opportunities but also forces both clients and translators to review their business models. For all those involved, it’s high time to redefine services, procedures and expectations, but most importantly to engage in a constructive dialogue with each other in order to find new ways of cooperation that suit everyone’s goals and needs.

Translators – one interlocutor in this dialogue – ought to eventually come out of their comfort zone and start seeing themselves as not simply linguists but as professional language experts with a wide set of skills. In the end, translation professionals have far more to offer than language knowledge and translational competence. For instance, they can provide deep insights into cultural diversity, have a strong understanding of quality, and can take on linguistic evaluations at different levels and in many applications.

I believe that the revolution our industry is currently undergoing offers a unique opportunity for translators to think about established services and possibly consider taking on new ones. Post-editing could turn out to be a clever addition to your own portfolio to offset the "quiet" weeks during the business year. On the other hand, translators could reach out to the clients who want to switch to post-editing in less suitable domains while expecting the same quality as a human translation. Here, translators could suggest a more translator-friendly use of technology. In fact, MT implementation in the CAT tool as a reference and not as an imposition could turn out to be an amazing productivity boost (what the client usually wants), whereby the linguist would keep control of the text and still ensure the expected quality.

In order to be on top of the game, translators should be:


  • open to new technologies and business opportunities, i.e. investing time in training and professional development, following the latest trends (not only in our industry), and reviewing goals and expectations on a regular basis;
  • proactive, i.e. reaching out to clients in order to establish a dialogue, talking openly about processes and quality, offering consultancy and explaining needs and expectations;
  • out and about, e.g. attending conferences and meetups; this is a valuable way of keeping in touch with both clients and colleagues, and of staying up to date.

Clients – the other interlocutor in the dialogue – depend on good post-editors in order to handle the ever-growing translation and localization volumes. Therefore, they should:


  • Make sensible use of technology and not just use MT for any kind of content assuming that it will always work well and contribute to saving time and money. Instead, they should learn to make conscious decisions and use the best approach. Post-editing might be a good option for specific texts or special requirements, while for more demanding content, it could be better to leave translators more room for creativity, either having them translating from scratch or offering MT as support, among other functions (augmented translation).
  • If they decide to go for MT, it is crucial to talk to vendors and explain the reasons behind the switch to MT or its implementation: Is it for time reasons, for cost reasons, or for both?
  • Clearly communicate the expected quality, providing specific guidelines with examples and checklists.
  • Offer post-editing training and feedback, especially to translators who have little or no experience with post-editing.
  • Allow ramp-ups for post-editing and possibly offer transitional rates so that translators can learn a new skill without significant revenue losses.

More generally, clients ought to reach out to translators exactly for the same reasons. For clients, it is very useful to find out about the processes and the work methods of their linguists, the issues and the concerns they might have while working on specific projects, the support they need and value, and finally some extra services they are able to provide.

I strongly believe that this kind of dialogue will foster better relationships between the different players in our industry and in turn, result in more and better business. Despite being unsettling, NTM could in fact turn out to be an exceptional opportunity, raising the entire sector to a new level. Now it’s up to each professional and each company to go and find their own place in this new landscape.