Machine translations (MT) is a hot topic now. I have seen and participated in several discussions in online translators’ forums on the topic recently. Translators and the industry as a whole are somewhat divided.
Many translators fear that the quality of machine translations is inferior to that of human translations and will ultimately drive down the overall quality and prices across the industry. Translation agencies feel increasing pressure to use machine translations to meet deadlines and constant budgetary constraints. Clients, who are aware of what machine translations are, feel that if MTs are being used in their projects without their knowledge, they are being ripped off. On the other hand, if machine translations are not being used at all, they are missing out on cost savings and faster turnaround times.
Machine translation is different from simply using a translation memory tool. Professional translators have been using computer-assisted translation tools (CAT tools) for decades to help them increase consistency and performance. The fundamental difference between the two is that in MT, a software will actually perform the translation and a human reviewer will then go over the translation editing it to ensure it is accurate and of good quality; this process is called post-editing of machine translation (PEMT). In human translations aided by CAT tools – which is the current norm across the industry -, memories and glossaries are created and available as reference to the translator. A software interface allows the translator to view the source document, similar segments already translated in the translation memory and other reference materials at the same time, while he/she types in translations for each segment. Despite all this “help”, ultimately, the translator carries out the actual translation of each segment of the source text. In both cases, each segment translated is stored in a translation memory (TM) and, in future translations, if the segment reappears, it is automatically input into the file being translated.
Hence, in the first process, there are usually two people involved, a post-editor and a proofreader, who will have the final responsibility for ensuring the quality of the translation. In the second process, there will be 2-3 people involved as well, the translator, a reviewer and sometimes a proofreader to ensure the quality of the translation.
Both of these processes sound great in theory. Ultimately, the translation will be reviewed and read by at least two qualified people, who will then ensure that the quality of the final product is the highest possible.
However, it is not that simple. Translation agencies are under a lot of pressure from clients to reduce prices. Hence, what happens is that these translation memories are used to “reduce the work” and the “cost” of the human assets involved. Translation agencies sometimes have grids for how much they pay for memory matches, for example, if there is a 100% match to a segment in a translation memory, the translation agency will not pay the translator for that segment. Therefore, the translator is not to touch that segment, even if its style, tone of voice, etc. are different from the rest of the translation, or worse, even if there are errors. If there is an 80-95% similarity between a new segment and a segment already in the memory, the translator may be paid 30% or 40% of his/her usual rate, etc. In other words, the translator will still have to read and edit that segment, but will not be paid in full for that service.
When CAT Tools and machine translations first came out, most clients were completely unaware of this, so often they would pay the full rate for all segments in a file, while translation agencies would profit from underpaying their translators. As clients caught on to this, they began pressuring translation agencies to reduce prices. This means that many translation agencies passed this pressure on to their translators, who were paid even less. No wonder translators are sceptical of new computer-assisted technologies!
Good translators eventually decided not accept this and would refuse to work with a grid or even with CAT tools. In response, translation agencies were forced to hire “cheaper” translator and the quality of translations across the industry spiralled down.
This is still an issue today, the middle and bottom of the translation industry are full of translators who work for a very low rate editing machine translations or translating 10-20% of a document without even looking at the rest to ensure consistency.
For the reasons above, I had been very sceptical of MT until recently. I only decided to try it, because I was asked by a client, happened to have the time and was curious about it. I completed three large projects for the same client, i.e. same terminology, same glossary, same style guide, etc. The first issue was that my client, a translation agency, was doing this at the request of their end client and not even they understood exactly how the rates were calculated. In other words, their clients had complete control over what they would charge and, hence, of what I would get paid (which was much less than my usual rate).
After my first project, I was actually excited about it. I did complete several words in a very short space of time. Therefore, I thought, maybe, I would eventually be able to earn as much per hour as I earn regularly. Hence, the lower rate would not be an issue.
It felt a bit like using a CAT tool, only all non-100% matches were like 70-90% matches and my job was to go over them editing. Now, 100% match segments all came locked, which means I could not amend them even if I did spot a mistake.
Nonetheless, there was still the quality issue. As translators are not paid for 100% matches, the client has to assume that we will not even read those. Therefore, they have to provide an extensive style guide and glossary to ensure that terms are translated consistently and the overall style of the translation is somewhat preserved. However, learning an extensive style guide takes time and even if you are really careful, when you are earning more than 50% less than you normally would for a job, you do not want to double the time it will take you to do that job by going over and over the style guide and glossary.
My first project was a short one, so it was relatively easy to ensure the quality of the segments I translated. It is very important to note here that I found several errors in the 100% matches, but I was told by my client (the translation agency) to simply ignore those. No one was getting paid for those so they were to remain unchanged.
In my next two projects, which were very large, I found myself under a lot of pressure and very confused! For example, I was supposed to be strictly faithful to the glossary, but sometimes I would be translating a segment and the term in the glossary would be different from how the same term had been translated in the 100% match segment just above. If I followed the glossary, as instructed, the text would be inconsistent. Why would you have two words to designate the same part of a machine in subsequent paragraphs? It is confusing!
In addition, there was the extensive style guide to follow, which I did not have time to learn fully, because I was supposed to turn the translation around quickly – after all, it had already been done by the machine, right? Well, the result and bitter lesson from this experienced was that these projects failed the “quality” evaluation during the proofreading stage due to consistency issues with glossary and style guide. I had not failed a quality assessment in at least 10 years! As a result, I had to amend and amend the translations until they were completely consistent with the glossary and style guide, which ultimately took me a lot longer than if I had translated the project for my full rate as I usually do. In other words, I lost money.
It was a bitter lesson, because having been in this industry for so long, I should have known that I could not have learned such a lengthy style guide so quickly, and should have factored this into my price and deadline. However, it has taught me a few things about machine translations.
Now, when I see articles like the one I read today, praising the “high quality, speed and cost savings of machine translations”. I take them with a pinch of salt.
My recent experience has not made me against machine translations. I do believe that there is value in them, because they can indeed speed up the process, but it has made me very aware that the way they are being done now, purely as a money saving exercise, is very detrimental to the overall quality of translations.
Ideally, for a machine translation to be of really good quality, the PEMT should review the entire text, not just non-100% matches. The reason for that is that a full revision will ensure that mistakes in previous translations are not perpetuated, and the tone of voice, style and terminology of the translation are consistent. Furthermore, the proofreader should be someone highly experienced with the style guide and glossary. Clients should treasure these professionals and the time investment they had to put into learning these style guide and glossaries, because they are the main quality element of the translation. These professionals should be paid more, not less, because they are “experts”.
The bottom line is that machine translations are excellent tools, like CAT tools, but they do not stand alone. Languages are living things that change and adapt, and no software today is able to adjust and account for that. It takes an experienced human eye to turn a machine translation into a good translation. The whole industry should be aware of this and properly reward these professionals, value their expertise and set procedures in place that allow them to do their job to the best of their ability. Only then will machine translations be “high quality, speedier and less expensive”.
This work is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported License.