Poll: Do you notice a drop in source text quality due to unedited, AI-generated drafts? Thread poster: ProZ.com Staff
|
|---|
This forum topic is for the discussion of the poll question "Do you notice a drop in source text quality due to unedited, AI-generated drafts?".
View the poll results »
| | | |
Especially mixtures of European Portuguese and Brazilian Portuguese, not to mention the misleading hallucinations…
| | | |
Lieven Malaise 
Belgium
Local time: 03:01
Member (2020)
French to Dutch
+ ...
Source text quality has always been variable and source texts that are not written by native speakers are hardly an exception. I might even argue that AI would do a better job in some cases.
Questions like these are not really useful (see also questions like "do you notice a drop of income due to AI?"). Who can really tell how big the impact of AI is? Nobody, in my humble opinion.
| | | |
| Not really. It would be great just to get some more source texts | Jun 3 |
Many of the jobs I used to do have probably been taken over by AI altogether.
I still see a few with a mix of sections written by different people. They are not primarily linguists, so native speakers of English do not necessarily write better introductions to their work than non-natives who have studied in the UK or USA for several years. Editing their English can be challenging!
I prefer to have the texts written in Danish, so I can translate them, but that can be a challen... See more Many of the jobs I used to do have probably been taken over by AI altogether.
I still see a few with a mix of sections written by different people. They are not primarily linguists, so native speakers of English do not necessarily write better introductions to their work than non-natives who have studied in the UK or USA for several years. Editing their English can be challenging!
I prefer to have the texts written in Danish, so I can translate them, but that can be a challenge too.
Some may actually be better with the help of AI, but they are definitely not unedited.
Other jobs are standard forms which were formulated without AI, which I translate and certify. ▲ Collapse
| | |
|
|
|
My sample size is too small to attempt any conclusions 😢😢
| | | |
| Maybe AI is particularly good at Flemish | Jun 3 |
Lieven Malaise wrote:
Questions like these are not really useful (see also questions like "do you notice a drop of income due to AI?"). Who can really tell how big the impact of AI is? Nobody, in my humble opinion.
Of course you can! Most of the time it’s still glaringly obvious when something has been written or drawn with AI.
The real trouble will come when it isn’t…
| | | |
Odd that there is no "No" option. I have not yet received source material that was obviously written by LLMs, or at least unedited. And like others above, an LLM would not make some of the errors I have had to deal with in human-written source texts over the years (though it would make others), so for some clients it would be an improvement!
| | | |
Lieven Malaise
Belgium
Local time: 03:01
Member (2020)
French to Dutch
+ ...
The Shredder wrote:
Maybe AI is particularly good at Flemish
The poll question is about source texts, so Flemish has got nothing to do with it in my case.
| | |
|
|
|
Lieven Malaise
Belgium
Local time: 03:01
Member (2020)
French to Dutch
+ ...
The Shredder wrote:
Of course you can! Most of the time it’s still glaringly obvious when something has been written or drawn with AI.
It's only glaringly obvious if it is used by people to have normal conversations or to express opinions, like on these forums and on social media. I don't translate normal conversations or opinions.
Everybody with half a brain would use AI only as a basis, not unedited. Just like any translator worthy of that name will use MT/AI only as a basis/draft. The problems with MT/AI begin where people start to use it to try to produce language they are actually not capable of.
| | | |
| Occasionally, but I see more indirect effects of AI on quality | Jun 3 |
I mainly work on clinical research and legal texts, which are less obvious victims of this (partly because privacy and confidentiality issues limit their use).
However, I see the quality of the overall text drop when all checking is left to translation software and nobody has time to ask if for instance the question and answers in a questionnaire align.
Since I am leaving translation and already get most of my income from other sources, I try to end my run in style. Often in ... See more I mainly work on clinical research and legal texts, which are less obvious victims of this (partly because privacy and confidentiality issues limit their use).
However, I see the quality of the overall text drop when all checking is left to translation software and nobody has time to ask if for instance the question and answers in a questionnaire align.
Since I am leaving translation and already get most of my income from other sources, I try to end my run in style. Often in a last read-through, I catch a glaring lack of deep reading of a text that has been through several revisions before they reach me.
Although this is more an effect of segmenting the text and seeing each segment in isolation (so older technology) AI has certainly made it more difficult to do a thorough job and still keep food on the table and a roof over your head.
[Edited at 2026-06-03 20:12 GMT] ▲ Collapse
| | | |
| MT target quality is definitely getting worse | Jun 5 |
I see a significant drop in MT output quality (i.e. target) in the last two years, for different language pairs. I'm not sure what is causing this and if this in any extent is related to the source quality also deterioriating. From my point of view the same MT engine just screws things up more often and in a more weird way. Why? For what it's worth, I don't care. All these moaning about "PEMT results being fed into the MT systmes and them getting better and killing our jobs" - give me a break. T... See more I see a significant drop in MT output quality (i.e. target) in the last two years, for different language pairs. I'm not sure what is causing this and if this in any extent is related to the source quality also deterioriating. From my point of view the same MT engine just screws things up more often and in a more weird way. Why? For what it's worth, I don't care. All these moaning about "PEMT results being fed into the MT systmes and them getting better and killing our jobs" - give me a break. This is not happening. It just churns out more unpredictable crap now. ▲ Collapse
| | | |
Lieven Malaise wrote:
It's only glaringly obvious if it is used by people to have normal conversations or to express opinions, like on these forums and on social media. I don't translate normal conversations or opinions.
Everybody with half a brain would use AI only as a basis, not unedited. Just like any translator worthy of that name will use MT/AI only as a basis/draft. The problems with MT/AI begin where people start to use it to try to produce language they are actually not capable of.
But then human nature takes over and they don’t bother correcting, or can’t see the problems. It’s a one-way spiral.
Maybe you don’t have the same exposure to English writing that I do, from social media right through to print journalism. It is still very obvious when it’s AI.
| | | |
Login to reply/comment 
