Bad Translations - AI or Deepl.com/Google Translate
I’m going to dictate this text as I often do, because typing for 12 hours straight is impossible. Sometimes the dictation goes well, other times it goes completely wrong and ends up full of the strangest words. For the translation, I use deepl.com. This translation app was recommended to me a few years ago here on this platform. Before that, I’d never even heard of it. Even back then, I noticed that the translator used completely different words to those I would use myself, and actually, this got worse and worse. The language is rather ‘silly’, as we’d say. A choice of words I wouldn’t use myself. Incidentally, this means there’s little benefit to be gained from the translation, as I have to check literally every word.
By now, we all know that AI is everywhere and it will be difficult not to make use of it. Most of us translate text on this platform, either because English is simply the lingua franca or because the community we’re posting in uses a language other than the one we’re fluent in. Most of us don’t speak more than one language anyway, and this is likely to become even less common in the future, as a translation app or programme is more convenient than learning a language yourself. On top of that, people are becoming increasingly less intelligent due to a lack of mental stimulation.
Now, a few years on, more and more people are using AI, and text is being corrected by it as well – so who knows what it actually says? I certainly don’t. It’s a service I provide to write for a number of people in their own language, because I know that translating English isn’t easy for everyone.
The fact that misunderstandings then arise is inevitable.
But how accurate are these AI text-checking programmes, really?
I’ve tried out a large number of them, and there are also websites that have a text checked afterwards by no fewer than 23 different AI text-checking programmes. The difference is shockingly large after I’ve recorded a text and later translated it. Whilst one says that 98 per cent was written by AI, another gives a result of 0 per cent. An average is then calculated from these figures.
Why does one say 98 per cent and the other zero per cent? The difference lies in what is being checked. Every language has its own specific word order (sentence structure) and grammar; such tools generally do not take this into account.
An additional option is to ask the AI whether it is possible that a text has been translated. If you do this, you’ll get a completely different result and may even be told what the difference is between a translation from Google Translate and one from deepl.com.
Will we ever be 100 per cent certain that something wasn’t written by AI? I don’t think so, unless you see someone typing the text out word for word and then pasting it into a translation programme, after which you see the result.
Is it possible to tell for yourself whether a text has been written by a human or by AI?
In my view, this is certainly possible if you pay attention to a number of points, alongside using common sense, being able to read well, of course, and taking the trouble to do some research.
I admit that Steemit still has very few good writers. Anyone accustomed to the typically poor content on this platform will undoubtedly find a good writer strange and quickly label this text as ‘AI-generated’.
I didn’t dictate the text above in its entirety, but started typing quite quickly because there are a few words that get misspelt when I say them out loud, as they aren’t recognised. And once I start typing, I just keep going...
Ik ga deze tekst inspreken zoals ik wel vaker doe, want 12 uur achter elkaar typen is geen doen. De ene keer werkt het inspreken goed de volgende keer wordt er niks van gebrouwen en staan er de gekste woorden. Wat de vertaling betreft gebruik ik deepl.com. Deze vertalingsapp werd mij een paar jaar terug aangeraden hier op dit platform. Daarvoor had ik er nog nooit van gehoord. Het viel mij toen al op dat de vertaler totaal andere woorden gebruikte dan die ik zelf zou gebruiken, en eigenlijk werd dit steeds erger. Het taalgebruik is nogal Popie Jopie zoals wij zouden zeggen. Een woordkeuze die ik zelf niet zou gebruiken. Dit betekent overigens dat er weinig winst valt te behalen op de vertaling omdat ik letterlijk elk woord moet nakijken.
Inmiddels weten wij allemaal wel dat AI overal te vinden is en het zal moeilijk zijn om daar niet gebruik van te maken. De meesten van ons vertalen tekst op dit platform, omdat nu eenmaal Engels de spreektaal is of omdat in een Community waar wordt gepost een andere taal wordt geschreven dan welke wij machtig zijn. De meesten van ons spreken sowieso geen meerdere talen en dit zal in de toekomst ook wel minder zijn, want een vertaal machientje of app is handiger dan zelf een taal leren. Daarbij wordt de mens steeds dommer door gebrek aan training van de hersens.
Nu een paar jaar later gebruiken steeds meer mensen AI en wordt tekst ook hierdoor gecorrigeerd en wie wat er dan staat? Ik in ieder geval niet. Het is een service van mij om een aantal mensen in eigen taal te schrijven, dit omdat ik weet dat het niet voor iedereen eenvoudig is om Engels te vertalen.
Het feit dat er dan misverstanden zijn valt niet te voorkomen.
Maar hoe precies zijn die AI text checkers eigenlijk?
Ik heb er een groot aantal uitgeprobeerd, en er zijn ook sites die een tekst later controleren door maar liefst 23 verschillende AI text checkers. Het verschil is schrikbarend hoog nadat ik een tekst heb ingesproken en later vertalen. Terwijl de een zegt dat het 98% door Ai geschreven is komt de ander met 0%. Vervolgens wordt hieruit het gemiddelde berekend.
Hoe komt het dat de een zegt 98% en de ander zegt nul procent? Het verschil zit het in wat er gecontroleerd wordt. Welke taal heeft zijn eigen specifieke woord volgorde (zinsbouw) en grammatica hier kijkt zo'n hulpmiddel in doorsnee niet naar.
Een extra optie is om AE te vragen of het mogelijk is dat een tekst vertaald is. Wie dat doet zal een heel andere uitslag krijgen en kan zelfs te horen krijgen wat het verschil is tussen de vertaling van Google Translate en deepl.com.
Zullen wij ooit 100% zeker zijn dat iets niet door AI geschreven is? Ik denk het niet behalve als je het iemand letterlijk ziet wrijven en vervolgens in een vertaalprogramma de tekst ziet plakken in het resultaat ziet.
Is het mogelijk om zelf te kunnen ontdekken of een tekst door een mens of AI is geschreven?
Volgens a u heeft dit wel degelijk mogelijk wanneer er op een aantal punten wordt geler, naast het gezonde verstand gebruiken en natuurlijk kunnen lezen en de moeite nemen om op onderzoek uit te gaan.
Ik geef toe dat Steemit nog maar weinig goede schrijvers heeft. Wie aan de doorsnee slechte content van dit platform is gewend zal beslist een goede schrijver tegenkomen bevreemdend vinden en al snel deze tekst het etiket AI gegenereerd geven.
Ik heb de tekst hierboven niet helemaal ingesproken, maar ben al snel gaan typen omdat er een aantal woorden zijn die verkeerd geschreven worden wanneer ik ze uitspreek omdat ze niet herkend worden. En eenmaal aan het typen, dan blijf ik typen...
If you use slop total to check, it uses 23 different tools and... The final results do vary quite a lot, though!
Original Dutch
12.9.26
Prompt: see title
#ccc #freewrite #ai #scb #translating #issues #steemexclusive
It is part of the future: either integrate or go back to papyrus. Although the sensation of smelling a freshly printed book is magical. Soon we will see more fakes, more AI integrated into everything. Everything is heading toward a massive lie—a point where we will eventually need copyists to hand-preserve whatever remains of art.
We artists will become oddities. Yet more necessary than ever in this cold world of screens.
0.00 SBD,
1.05 STEEM,
1.05 SP
At least we artist will be rare and therefore special. I believe papyrus and rice paper it totally fine. We can go through process of making our own paper, ink, paint and whatever we need or we write in the sand, paint walls and roofs. If needed our skin.
I will not be around if that period of AI taking over, what @lupega fears, will be normal.
I will be no bionic woman, not one million or six or twelve. It is a relief.
0.00 SBD,
0.19 STEEM,
0.19 SP
Maybe being a bionic woman has its perks. Like being able to deliver a solid whack to the head. Or making a dog roll around in distress with an ultrasonic sound when the neighbor and her mutt are being a nuisance. You could turn on the TV or hack NASA just by thinking about it, haha.
I agree with you 100% because the translator often alters the words we intend to use, changing their meaning and preventing us from conveying our message accurately. I have noticed this myself: whenever I translate my own posts, the wording and phrasing often shift; similarly, when I read others' posts—which I also translate to understand—I often fail to grasp the full meaning, especially since the original author likely used a translator as well. This disconnect makes it difficult for us to truly understand one another. I usually try to avoid commenting in such situations, but a few days ago, I made a mistake that I still regret; I commented on someone's post, and due to a translation error, my comment was interpreted incorrectly.
0.00 SBD,
1.03 STEEM,
1.03 SP
That indeed is the point. It alters and plays with words as if we were playing some sort of word game.
If you refer to your remarks based on observations, that was no translate mistake. The translation was good and you were right.
You were not the only one who noticed it. Actually, you were kind. All you said was that you can see the two of them are close.
Why that tirade directed at you is still unclear to me. It has nothing to do with you or the translation, but the one who responded.
If we are not sure what is said, we can ignore it or simply ask what is meant. I am sure most of us know we all, reading international Steemians, have to translate.
Thanks for sharing your experiences.
🍀❤️
0.00 SBD,
0.13 STEEM,
0.13 SP
That is exactly the case; we have to rely on translation to convey our message. Since that incident, I have made it a point to read through every paragraph I translate, and whenever I spot a questionable word, I try to replace it with the correct one.
Friend, you can also ask what is meant and say you are not sure about the translation.
👍
I have not used deepl before. I generally use AI a lot but mainly for preparing proposals, reword my work emails and such. But never for translation although am not sure if Google Translator uses AI in its translation. And slop is another tool I have not heard before. Interesting.
But when AI is used, I usually based the usage on its perfectness in the choice of words. This is in relation to emails mainly, especially when I know the person or if I have spoken with the person before, and I know the level of spoken English of that person is not the level that I read in email.
But when it comes to AI translation, I suppose we can use the same logic? If it reads too perfect, it does look AI generated? But can we blame them if English is not their spoken language? When I translate to Spanish, I have no idea if the translation is correct and can only trust that the translator is doing its job properly.
0.00 SBD,
0.27 STEEM,
0.27 SP
It’s not about whether what’s written is perfect Spanish, English or any other language, but rather that the translation app or AI uses certain word choices that the average Steemian simply wouldn’t use. This also applies to a number of standard phrases, and from these you can clearly see that the text wasn’t written by a human. These kinds of standard phrases are particularly noticeable in stories.
To be honest, I wouldn’t know what I’d actually need AI for, apart from generating images – something I learnt on Steemit. The Italian community organised a competition where you had to do just that.
Translation apps and programmes will also make use of this, just like Grammarly, Canva, the standard search engines in browsers and who knows what else.
0.00 SBD,
0.13 STEEM,
0.13 SP