The Reluctant Translator
Click here to close
Click here to close
Subscribe here

The Reluctant Translator

The Reluctant Translator

The Reluctant Translator

Neural Machine Translation (NMT), Adaptive MT, Forward-Thinking, Advanced Leveraging, Zero Shot Translation…all these new technologies made me wonder what our translators are thinking. They are the ones who should be most concerned about tools that can impact their daily work. Are they using MT? If so, where and when? How is their productivity impacted?

So, I sent an email to my friend—a long-time, experienced translator—with my questions about how his world has changed thanks to translation technology. His response brought up a lot of interesting points.

Payment is tricky

Payment to translators has evolved along with the quality of MT. The typical process is this: first, Translation Memory is applied to the text; then, remaining content is sent through MT. In the old days, when the quality of MT wasn’t great, translators used the machine translated content as a suggestion; they could take it or leave it. As a result, they were paid in full for those words.

But since then, the quality of MT has improved, and these segments are now only paid at about 60% because the translator “has less work to do”.

(It may or may not actually be less work to translate using MT: many translators consider post-editing to take just as long as full human translation. For more on this debate, check out our blog post MT Post-editing Myths Debunked by an Expert.)

Tracking is difficult

With CAT tools, files, and TMs moving into the cloud, work is dynamically spread across multiple translators. So, tracking who gets assigned how many repetitions and what the leverage looks like per person becomes difficult. One translator may be getting the majority of the repetitions, while another may have to do all the translation from scratch. Quite often, translators can’t anticipate how many net words they have to translate. Payments and schedules are now tricky to estimate, and LSPs have to sift through a lot more wordcount data to figure it all out.

Quality is still variable

MT quality has improved, but there are still substantial errors which can be easily overlooked. For example, a sentence may not be properly negated when the MT engine misses a single word such as ‘no’ or does not distinguish between ‘can’ and ‘can’t’. (Single-word errors are even more common with Neural MT engines.)

There are problems with consistent use of terminology also. It’s possible that an MT engine might not understand when to apply domain-specific terminology, or may ignore a given glossary altogether. Also, a translator may need to correct terminology during a job, but that correction isn’t implemented in the engine; the translator has to remember to make that correction over and over.

And lastly, MT might not be good enough at all. If MT outputs are not great, and post-editing can’t bring the translation up to the desired quality level, a professional and experienced translator ends up translating the task from scratch faster and with less risk.

Tools change…and not always for the better

Constantly changing tools and frequent version updates are another problem. Productivity is a very important factor for a professional translator. Time really is money. The tools and environment need to work like a well-oiled machine. Just a simple change of a hotkey will result in a lot of cursing and potentially broken keyboards. Beta versions or even adopting a brand-new release of an application right away should be avoided. These are usually buggy and may easily result in a loss of data.

This all sounds quite negative, and it’s understandable that translators may feel reluctant about technical progress (hence the title of this post). Seasoned translators may have a different attitude, depending on their languages, domain, and personality, but regardless, they will definitely be able to relate to these concerns.

Expanding the discussion

To get broad opinions on translator attitudes towards MT and CAT tool improvements, we conducted an anonymous survey among our translation suppliers, SLVs, MLVs, and freelancers.

Guess what? There is quite a mix of opinions about MT and tools. And as you may have guessed, there is also a correlation between experience and acceptance. Younger folks, new to the translator profession, are more likely use MT to increase their productivity—and even enjoy doing so. More veteran professionals may be a bit more, shall we say, stuck in the mud.

But we found much more. I’m teasing you with what I’ve already described here, I know. But I’m not going to throw the data out there; rather, I want to give various stakeholders from the industry the chance to discuss and analyze the findings.

And on what better platform than LocWorld?

On November 3rd at LocWorld35, I will meet with Spence Green (CEO and co-founder of Lilt), Jost Zetzsche (Translator and Consultant at International Writers’ Group), and Paola Estrella (NLP, MT, and PE expert at Moravia) in Santa Clara for a panel discussion. We will review the main findings of our survey, discuss options to address translator concerns, and brainstorm how to increase acceptance of MT and tool improvements.

If you can’t join us, please look out for a future blog post where I’ll summarize our findings.