Trends in the language services industry: Machine translation, part 1

 

Roughly 80% of translators report they are using MT in 2022. To be clear, this refers to the application of MT in instances in which the translator decides that it will be beneficial in some way; we are not necessarily talking about running full jobs through MT, although in some cases that can also be an option. For the most part, MT applied by translators happens in specific cases, and is used in a variety of specific ways. It may be applied more in some jobs than in others. It can be a time-saver on simple texts or lists. It may be used as one more set of possibilities from which to choose in the CAT tool they are using. Stuck on a term? Your MT may give you the right result quickly, or the incorrect result it gives helps your brain make that click that provides you with the right translation. Each translator uses it in a way that works for them and what they are working on. An increasing number of translators are finding that there are advantages in doing their own machine translation post-editing as well, where the text is run through MT and the results are refined where needed by the translator. There are some caveats to this, which we will look at below.

 

Quotes from language professionals:

Having been a translator for more than 30 years, I can say that technology has streamlined my work and increased my productivity and earnings multiples times over. I remember when terminology search meant going to a library. Because I am a technical translator, that meant travelling to the city where the university was physically located...
 

 

Quotes from language professionals:

A client specifically asked me to post-edit a translation for a computer catalogue and was furious with the result being too close to the original.
 

 

This is the use of MT which we may have heard most about, or which most easily comes to mind when someone starts talking about machine translation, as it has tended to cause the strongest reaction among language professionals. One frequent complaint here is the use of machine translation to further push down rates for translation work.

 

To be clear, this usage is not bad in of itself. For example, a client may have varying quality needs in the content they want translated. In some cases, the quality required is “just good enough to understand”, and the MT solution is such that it would be less cost effective and more time consuming to involve a professional human in part or all of that particular project. Knowing when and where and how to apply a technical solution like MT to improve results is the smart thing to do. And there are companies or clients who are making smart use of MT solutions. We tend to hear more about the cases that end in disaster or tragedy, in part because those make for better or funnier news items.

 

One piece of news you may have seen in late 2021 on the use of MT was when the Spanish Association of Audiovisual Translation and Adaptation, ATRAE (Asociación de Traducción y Adaptación Audiovisual de España), published a statement decrying the alleged use of machine translation post-editing (MTPE) by Netflix for their hit Korean series Squid Game. The statement blanket asserts that post-editing leads to lower quality results and also focuses on the generally lower rates offered for many post-editing jobs. But, here is where it gets interesting: while the statement got some traction initially in Spanish, it appears that a human translation to other languages of the statement or its related news items was not immediately forthcoming. So, AI and MT stepped forward, to dispassionately help spread the word. The posts bringing this news to English, for example, were not post-edited, but were apparently directly AI-translated with machine translation, and this is how most of the world heard about the objection being raised, with phrases such as...  

 

• Bread (hard and dry) for today, hunger for tomorrow.
• ‘The squid game’ is the fiction of the moment. He has shattered the record for the most viewed series of Netflix so far in the hands of ‘The Bridgertons’ and their popularity seems to have no ceiling.
• Now it has been the Spanish translators who have raised their voices against her.
• [On what post-editing is] Here you can read a twitter thread that explains in detail the process and why its boom is due.

 

It is not clear what impact on public perception this had, a news item about translators against machine translation which was itself poorly-translated with machine translation. It is possible that the general reader, unfamiliar with how and where MT is applied, reads a text like this as simply poorly-written by a human, rather than poorly-translated by a machine, with the resulting possible bias that can come from reading a text in your language which is hilariously inadequate.

 

At any rate, back to machine translation applied by clients. Where problems seem to arise is when the appropriate communication surrounding the job to be done, or the appropriate preparation to carry out a job that involves MT, or both, are missing.

 

How this presents itself varies. Let’s take a look at some brief example cases that were frequently brought up:

 

1. The translator is given a text that has already been passed through MT and asked to perform proofreading. In some cases, the text is presented as having been translated by a human translator. At times, no source text is offered. In instances where the MT result is good, the translator may feel uncomfortable making only minor changes, or the client may be upset that the end result is “too close to the MT result”.
    
2. The translator receives a text pre-translated with MT, the quality of which is so low that in order to produce a result that meets the quality needs set, the text must be translated “from scratch”, adding additional time to the project.
    
3. The translator submits a finished translation, and the client checks the results using MT, adding a layer of confusion to the proofreading or quality assurance process.
    
4. The translator is given a project where segments have been filled in with fuzzy matches, with the understanding that these will be discounted, and told these are from a TM, when they are in fact MT.
    
5. The translator is given a poorly-translated text which must then be translated to other languages. For example, a text originally written or generated in Chinese, MT-translated to English, and the translator is asked to translate the English result to a third language.

 

Translators report that some agencies they work with have introduced a nearly blanket application of machine translation post-editing for their services, only to walk this back when the discount on the service and the quality of results are not a win-win for the client. In some cases, the smart application of MT works in favor of the human translator who can then command a slightly higher rate for those jobs where their brain and skills are a must, and the agency is able to pay that rate and still make more by compensating with discounted MT-related services on the other end. In other words, adding MT-related services as an option to clients can allow a translation company to pay their human translators more, and still make more themselves.

 

A recurring complaint among translators when some of these MT solutions are thrust upon them is the dehumanizing effect it can have when adequate care is not taken in their implementation. Timed post-editing tests for candidates to a project sound like a good idea on paper, but may not have a positive effect on the professional vying for that work.

 

And what happens when you are fed to the machine and it chews you up and spits you out? At least one translator relates the story of having been an employee at a company where the translators spent several months training and refining a new MT engine. At the end of that project, the translators were let go.