“They get what they’ve paid for”

anonymous person holding coins
Photo by RDNE Stock project on Pexels.com

An intriguing discussion among translators recently caught my attention. It revolved around a client’s attempt to cut costs by pre-translating texts using machines, followed by offering the job for editing at an extremely low rate. The consensus among translators was clear: such underpaid tasks might not receive the attention and effort they require.

Savings Over Quality?

It’s no surprise that many clients turn to machine translation to save money. The allure of using machines instead of human translators, with the promise of later editing, is tempting. Indeed, machine translation post-editing (MTPE) has become a frequented workflow, occupying a significant portion of translation jobs in various market segments. But is this approach flawless?

Quality vs. Cost

Machine translation (MT) providers often assert that their output quality is nearly perfect. Undoubtedly, these machines have made astonishing advancements. But here’s the catch: while machines have improved, it’s crucial to align your expectations with what they offer. If your priority is saving costs rather than attaining flawlessness or creative work, using machines might suit your needs. Just don’t expect miracles.

But you’ll have it edited, so it’s going to be perfect?

Here lies the crux of the matter. The translation market faces a significant gap between the expected quality of machine-translated output, a post-editor’s capabilities, and fair compensation for the work.

Contrary to claims by some providers that MTPE should rival human translation, the reality is quite different.

Post-editors aren’t tasked with crafting original translations. Their role involves editing machine-generated content to ensure coherence, grammatical accuracy, contextual consistency, and adherence to prescribed terminology.

  • The focus isn’t on creating the ideal translation; it’s adapting the machine’s work to the minimum required extent.
  • The market trend of lower rates for MTPE results in post-editors handling higher word counts per hour, leaving less time for refining terminology or style.
  • There’s no room for terminological research. Pressed by time, inexperienced post-editors are tempted to use the MT output without properly checking, often with a painful result.

The real savings from MTPE

Determining the actual savings from MTPE isn’t straightforward. The effectiveness of over 30 machine translation engines, including 3 AI-powered ones, varies based on content domain and language combination.

Are clients the most qualified to decide whether to employ MT and select the best provider?

Remember: A language professional will probably be more competent about this decision. And I am not sure whether the fact that if you proceed to this step on your side anyway, justifies any expectations of significant discounts from translators. Would anyone give you a 50% discount just because you’ve done a 2-Dollar step for them? Maybe also not in the best manner?

Insights from Translators

11% translators agreed that they have to work faster on MTPE jobs to achieve the same hourly compensation (compared to translation), assuming the client expects lower quality.

14% state that according to their experience, MT output is unusable. (7% usually delete MT and retranslate from scratch, 7% won’t accept MTPE jobs any more)

43% are OK with MTPE, stating their hourly income on MTPE is about the same as from translations, but elaborating on this response, they provided some game-changing comments:

Language pros satisfied with their compensation for MTPE:

  • Charge all their work by hour; or
  • Offer only negligible discounts on MTPE, or none; or
  • Process high numbers of words per day, claiming the quality remains perfect.
    Note: My experience of a senior reviewer however, contradicts such statements:
    in many cases, the quality is substandard. It seems to be clear that post-editors are forced to produce at high speed even if the MT quality is not good for their field and language combination.
  • MT usability is highly dependent on domain and language combination.

“…I do see the quality dropping quite a lot due to MTPE: partially it’s because linguists are struggling to maintain their same earnings in terms of time and so skip through docs… but it’s also due to less experienced linguists being given MTPE jobs because there is a belief that there is “less to do” with the document but also because the less experienced linguists are happy to accept lower rates…”

“I went from 2500 source words per day to around 4500-5000 by using MT as an aid.”

“In my IT>EN language pair for technical pharmaceutical translations, the MT output is actually very good. … This is not the case, however, for my ES>EN language pair and legal material.”

“When I started doing MTPE 4 years ago, I made a point to calculate my MTPE rate … to maintain my average hourly rate. many people are starting to accept MTPE jobs for less than I do, and my work volume has started to decline…”

“First of all, my hourly rate for MTPE is a good rate for the majority of my MTPE projects, so I’m ok with this rate and I’d say I finish 90% of my projects within the time allocated.”

Many have stated, some very roughly, that using MT makes no sense. These were often creative translators. But this shouldn’t come as a surprise.

Choosing the path for your project

Ultimately, the choice is yours. Do you seek creativity or simple editing? Superior quality or cost-effectiveness? Are you a tech enthusiast eager to explore all that technology offers? If so, get advice from professional linguists for optimal results.

How about using MT only?

Like no editor at all? Honestly, I saw projects where the editor’s contribution to quality was negligible. This, however, is not to say the text was OK. I mean that the result remained disastrous.

You CAN use the machine translation directly. Just consider the purpose: Is it permissible if the result suffers from omissions, additions, shifted meaning, a nonsense here or there, or pure gibberish?

Are you able to assess the usability of the text in the target language?

Leave a Reply

%d bloggers like this: