| Pages in topic: [1 2] > |
Poll: Would labeling translations as "human-verified" improve market quality? Thread poster: ProZ.com Staff
|
|---|
This forum topic is for the discussion of the poll question "Would labeling translations as "human-verified" improve market quality?".
View the poll results »
| | | |
| What Should “Human-Verified” Actually Mean? | Aug 28 |
I asked AI to look into whether this topic has been discussed elsewhere, and I found the results quite interesting. It seems that the idea behind this poll is actually part of a much broader discussion in the translation industry.
For example, FIT (the International Federation of Translators) has discussed the idea of distinguishing “Professionally Verified Translation” from unverified translation, particularly in the context of AI and machine translation. The underlying idea is... See more I asked AI to look into whether this topic has been discussed elsewhere, and I found the results quite interesting. It seems that the idea behind this poll is actually part of a much broader discussion in the translation industry.
For example, FIT (the International Federation of Translators) has discussed the idea of distinguishing “Professionally Verified Translation” from unverified translation, particularly in the context of AI and machine translation. The underlying idea is that users should be able to understand whether they are receiving a professionally verified translation or simply an unverified output generated by a machine or AI.
This is interesting because it is very close to the question being asked in this poll. In other words, the concept of making the human contribution visible is already being discussed at the level of an international professional organization, rather than being merely an idea among individual translators.
At the commercial level, some translation and localization companies are already using similar terminology. For example, some companies market services using terms such as “human-verified AI translation”, meaning that an AI or machine translation is produced first and then reviewed or validated by a human linguist. Other services have gone even further and use “Human Verified” as a specific service or certification level, particularly for documents that may need professional or official verification.
There is also increasing academic interest in the question of how much human involvement is necessary when AI is used for translation. Recent research has begun looking at different levels of human involvement, ranging from AI-only translation to workflows involving increasing levels of human review and verification. Some studies also suggest that while AI can perform relatively well in areas such as basic meaning and terminology, human involvement can remain particularly important for things such as clarity, grammar, cultural appropriateness, and overall quality.
So, from what I found, this does not appear to be simply a question of “AI vs. human translation.” A more interesting question may be how much human involvement is actually present, and whether that involvement is meaningful enough to justify a particular quality label.
And this brings me to what I think may be the biggest problem with the term “human-verified.”
What exactly does it mean?
Suppose Company A does this:
AI translation → human quickly skims the text → clicks Approve
while Company B does this:
AI translation → professional translator reviews the entire text → terminology check → bilingual review → QA → final human approval
Both companies might potentially describe their service as “human-verified,” even though the amount and quality of human involvement are completely different.
This makes me wonder whether a “human-verified” label would actually improve the market unless there were some kind of agreed definition or standard behind it.
Perhaps we would eventually need something more along the lines of different levels, for example:
AI-generated — AI-generated with no human review.
AI-assisted — AI was used as a tool, but a human translator performed the translation.
Human-reviewed — an AI/MT output was reviewed by a human.
Human-verified — a qualified human linguist reviewed and formally took responsibility for the quality according to defined criteria.
Professionally certified — the translation was reviewed or certified according to a recognized professional or regulatory standard.
Of course, these are just examples, and I am not suggesting that these particular definitions should become an industry standard.
But I think this raises a much more fundamental question: Could a “human-verified” label actually help restore some transparency to the translation market, or would it simply become another vague marketing term?
Personally, I can see a potential benefit. If clients increasingly encounter AI-generated translations, they may eventually want some way of distinguishing between an unverified machine output and a translation for which a professional human has actually taken responsibility.
On the other hand, if anyone can put a “Human Verified” label on a translation simply because a human looked at it for a few minutes, the label could quickly become meaningless — or even make the situation worse by giving clients a false sense of quality.
So perhaps the real question is not simply:
“Should translations be labeled ‘human-verified’?”
but rather:
“If we introduce such a label, what should the label actually guarantee?”
I would be very interested to hear how other translators, reviewers, agencies, and clients see this. Would you trust a “human-verified” label if there were no clearly defined standards behind it? And if standards were introduced, what should the minimum requirements be? ▲ Collapse
| | | |
Zea_Mays 
Италия
Local time: 00:24
English to German
+ ...
Human translation is the easier solution, and some businesses already reverted back to it (personal experience).
| | | |
I have been doing human-editing for some of my customers who decided to translate their documents with DeepL, but I doubt they would ever want their texts labelled…
| | |
|
|
|
Gregor Trebec
Словения
Local time: 00:24
English to Slovenian
+ ...
| Doesn't mean nothing | Aug 28 |
It just means it was read and understood by a human but the real problem persists.
| | | |
Daryo
Local time: 23:24
Serbian to English
+ ...
| What exactly 'market quality' is supposed to mean? | Aug 28 |
A make AI generated output easier to sell as 'real translation'?
B improve the intrinsic quality of what is being sold?
Any educated guess or by some miracle an actual explanation from whoever asked the question?
| | | |
"human-produced"? 
| | | |
I’m trying to figure out what "market quality" means. Does it refer to the quality of the translation, the image of the translation industry, or the client's positive reception/perception?
In any case, I don't think that will improve anything. Quite the opposite: it is actually a more subtle way of suggesting that your translation wasn't done by a human.
It is a bit like the "Designed in France" label that some French retailers use to attract customers' attention, yet... See more I’m trying to figure out what "market quality" means. Does it refer to the quality of the translation, the image of the translation industry, or the client's positive reception/perception?
In any case, I don't think that will improve anything. Quite the opposite: it is actually a more subtle way of suggesting that your translation wasn't done by a human.
It is a bit like the "Designed in France" label that some French retailers use to attract customers' attention, yet this kind of lable could actually achieve the opposite result: in reality, the label implies that the product is not actually manufactured in France - that's how some clients interpret it. And clients are right. ▲ Collapse
| | |
|
|
|
Daryo
Local time: 23:24
Serbian to English
+ ...
| Cloud cuckoo-land? | Aug 28 |
Puritad Thongpreecha wrote:
...
On the other hand, if anyone can put a “Human Verified” label on a translation simply because a human looked at it for a few minutes, the label could quickly become meaningless — or even make the situation worse by giving clients a false sense of quality.
...
Given the economics of this whole affair what would you expect to happen?
Short version: making really sure that the AI output is as good as a translation by a competent human is bound to take MORE time and cost MORE that a real translation by a competent human.
The only way to make it cheaper than real translation by a competent human is to employ people who - even if capable of - given the rates paid to them could simply not afford to take the time necessary to check in-depth the AI output.
In effect they'll just be used to take the blame if anything goes wrong.
Believing the opposite is living in Cloud cuckoo-land about as much as those claiming that CAT tools / MT / AI was pushed on everyone so that translators could earn more.
| | | |
Lieven Malaise
Бельгия
Local time: 00:24
Member (2020)
French to Dutch
+ ...
Daryo wrote:
Short version: making really sure that the AI output is as good as a translation by a competent human is bound to take MORE time and cost MORE that a real translation by a competent human.
I'm sorry to read that you are not able to process machine translation properly and profitably. That doesn't mean that others can't, though, so your opinion or limited observation shouldn't be considered as some kind of general fact.
Daryo wrote:
The only way to make it cheaper than real translation by a competent human is to employ people who - even if capable of - given the rates paid to them could simply not afford to take the time necessary to check in-depth the AI output.
Profitable MTPE rates can be negotiated, in which case there is all the time necessary to check the machine translation properly. Not everyone works for a pittance and not every agency or customer expects you to. Being succesful in MTPE is a combination of skill acquired through training/experience - making it possible to lower your rate reasonably - and maintaining decent rates instead of giving in to shameless bottom-feeders.
| | | |
| Short (translation) memories | Aug 28 |
I can remember (because I had to review it) some of the absolute nonsense produced by humans back before MT was widespread. "Human" is no guarantee of quality whatsoever.
In fact, it probably goes some way to explain the poor quality of MT/AI translation as it stands.
| | | |
Charlie Bavington wrote:
"Human" is no guarantee of quality whatsoever.
Even if a human has actually done the verification.
Anybody can stick a label on something, saying anything they wish. It doesn't mean it's true.
In any case, this discussion is academic, because within a few years AI translation will be better than human translation in all but a tiny minority of cases. (And of course, faster and cheaper.)
A better question would be: how long will it take before clients insist on their translations being certified as "AI verified"?
| | |
|
|
|
| Human involvement means nowt, as does “human-verified”. | Aug 29 |
There are plenty of very competent human translators, but alas they are greatly outnumbered by the not-so-competent ones. Anyone can claim to be a translator. Yet they’re all "human".
MT/AI outperforming the majority of humans is not as impressive as it might sound. And having a human involved in a translation process doesn’t mean much either. Hardly an automatic stamp of approval.
| | | |
PiotrSz
Польша
Local time: 00:24
French to Polish
+ ...
Philip Lees wrote:
In any case, this discussion is academic, because within a few years AI translation will be better than human translation in all but a tiny minority of cases. (And of course, faster and cheaper.)
That's what owners of AI companies want people to believe but it's probably never going to happen and certainly not "within a few years".
[Edited at 2026-08-31 16:43 GMT]
| | | |
Gregor Trebec
Словения
Local time: 00:24
English to Slovenian
+ ...
Only 'thing' that would help is signing a clause or an agreement not to use AI.
| | | |
| Pages in topic: [1 2] > |
Login to reply/comment 
