In a world increasingly flooded with AI-generated content, have we forgotten how to read critically?
A thought-provoking piece I read recently entitled “Promptism and the Forgotten Art of Source Criticism in the Age of AI” raises exactly that point.
It made me think about legal translation.
And how what it says about source criticism applies so powerfully to what we do as legal translators.
You’ve probably seen it too.
Legal texts fed through AI tools that come out sounding smooth, plausible, grammatically correct.
But as we all know (or should know ) legal translation isn’t about sounding smooth.
It’s about being right and precise.
The article talks about the illusion of fluency; how something that reads well can still be deeply flawed.
And in legal settings, flawed translations mean flawed understanding.
And flawed understanding can lead to real-world consequences.
When working on legal texts, especially ones involving rights, obligations, or procedural safeguards, “close enough” is not close enough.
This is where source criticism (the habit of actually going back to check, verify, and compare with the authoritative original) becomes a professional duty.
Not just a nice-to-have.
If you’ve ever run an AI-generated translation through its paces, you’ll know it rarely holds up under scrutiny.
Subtle distinctions missed.
References mistranslated.
Wording altered in a way that changes legal effect.
It doesn’t matter how confident or elegant the output appears. If it isn’t anchored in legal reality, it’s unusable.
And more than unusable, it’s potentially harmful.
That’s why I keep championing the role of expert human translators.
Because we legal translators don’t just “translate”. We verify, interrogate, and challenge the text.
We bring legal knowledge to the task.
We bring linguistic precision.
We do the checking so our clients don’t have to.
The message of that Promptism article is clear: don’t lose the habit of critical reading.
Don’t stop questioning what you’re shown.
In legal translation, that habit can and does make all the difference.
If you need Greek to English legal translations that go beyond the surface and actually hold up in practice, get in touch with us at #Jurtrans. Because close enough simply isn’t good enough.
Following on from yesterday’s musings on the utility of technological tools for legal translation purposes, let’s hear today what a lecturer in family law has to say about the topic:
“Regarding … improving family experiences, numerous AI-based tools are designed to support clients or lawyers, but their accuracy strongly depends on the case details.
No doubt automatised legal translators [sic] have rapidly progressed in the last few years, yet their accuracy compared to the sworn translator can still be questioned. …
Yet, they tend to fall short in complicated matters, particularly highly contextual sentences.
Additionally, as a scholar deeply immersed in Japanese family law, I can assess that the existing translators [sic] can mislead about the true sense of the content of the legal norms or documents and, thus, can be treated only as a support tool. …
Family law requires an exact understanding of the analysed text, including legal terms and human emotions that can be expressed in various ways, such as non-verbal messages. …
Given the noticeable number of international couples communicating with each other or their children in different languages, detailed knowledge about the family and personal situation cannot depend on automatic translation, which cannot grasp the significance of words and non-verbal messages at this stage.
Conversely, clients seeking legal information through automated translations could find incomplete or false information.
Thus, despite the rapid pace of development of AI-based legal translations, family lawyers should consider employing reasoning based on them as high-risk assumptions, mainly due to the inability to grasp subtleties and cultural nuances.
This raises an essential question – is AI-based translation useful at all if a specialised human translator will probably always be more accurate than machines and one step ahead, despite the slower work?”
Piegzik, M.A., “The Adoption of Artificial Intelligence in Family Law – Brand New or Well-known Idea?” Folia Iuridica Universitatis Wratislaviensis 2024 vol. 13 (2), 26–51
I’d argue these considerations do not apply just to family law but to all branches of law.
The language of the law is hard. Translating the language of the law is hard.
It still confounds technology. There are no simple push-of-a-button solutions.
If you need help seeing through the hype and want someone to tell you the truth about what language technologies can and cannot do when it comes to translating legal documents from Greek to English and vice versa, book a 30-minute call here: https://lnkd.in/edMThP3c
I’m all for technology. I use it all day, every day. But every technology has its limits (and its limitations).
Don’t we all wish that legal language was simple and straightforward and that you could just click a button and you’d have a clear understanding of what a foreign legal text actually says?
Then you’d know what your rights or obligations are.
You’d be able to converse intelligently with your legal counsel or your lawyer could explain things to you in terms you understand.
You’d be able to plan your next steps.
You’d know where you stand.
Nonetheless, 9 years after NMT went mainstream, and 2+ years after genAI was launched, legal language continues to confound technology.
By way of illustration, let’s take a sentence that would be straightforward for any Greek lawyer or experienced professional legal translator working in the Greek-English combination:
The Greek source text
The sentence may be straightforward. But behind it is someone who needs to show they have not been accused of something serious. The words may look innocent on the screen but the real-life implications are huge.
It’s vital to get the translation right.
You don’t want just any old translation.
You want an impeccable translation you can rely on.
How the tech performed
Google Translate gives us this piece of incomprehensible garbage. Remember Google Translate has been working on fine-tuning its neural machine translation engines for 9 years now.
The “English” translation according to Google Translate
– This is neither fluent nor accurate.
– Is employee even correct here? Is the text not actually referring to a “civil servant”?
– What does “occupied the service” even mean?
– The employee occupied the service “due to disciplinary or criminal offences”. What does that even mean?
– Holiday? αργία can mean public/official holiday yes, but in a completely different context. This is a bit like opening a dictionary, seeing 10 different meanings of a word, and picking the first. It might be correct in some context. Not this one.
In law, context is king.
DeepL gives us this rendering:
The “English” translation according to DeepL
– Again this is neither fluent nor accurate.
– Again we have to ask whether employee is the right choice of term here.
– Why is the “suspended from duty” idea now in the first part of the sentence when the Greek is referring to the fact that the department/unit the person works for has not had to handle any disciplinary/criminal complaints against that person?
– “Placed on leave”. Does that fully convey the concept of αργία? Or is it only covering part of the concept? αργία/argia means you are (i) stripped of the right to perform your duties + (ii) you only receive half your salary. “Placed on leave” really doesn’t cut it.
– Placed on “suspension from duty”. Grammatically does that stand in English? It’s beyond awkward and definitely doesn’t flow. Let’s say αναστολή here does mean that the person is suspended from duty. Is that with pay or without pay? How is this concept related to the previous one (αργία)? How does the “translation” differentiate between the two so that the reader actually understands what is going on?
Chatgpt 4.0gives us this:
The “English” translation according to Chatgpt
– This is again not fluent English. Nor is it accurate.
– As with the other versions we have to ask whether “employee” is the correct term.
– The first half of the sentence sort of conveys the meaning but omits quite an important bit of information (that the place where this person works has not had to handle complaints against them). Let’s say you get some general idea. In legal settings, do you really only want to walk away with “some general idea”?
– The second half of the sentence doesn’t fluently link to the second half.
– If αναστολή is the concept of “suspended” why does Chatgpt render αργία as “placed on suspension”? Does it draw an adequate distinction between the terms? From the viewpoint of Greek law, the two terms are quite distinct (related yes, but distinct). Let’s say there is an element of “suspension” in αργία, does the English rendering this tool offers accurately and fully convey what the Greek concept is saying? Or does it not omit a huge part of the meaning? What of the point we highlighted above concerning reduction in pay? How does that get conveyed by the wording chosen here?
Reality check
If you’re an ordinary individual using these tools to get some idea of what the Greek text is saying, can you hand-on-heart say you really understand the Greek after reading these “English” renderings?
If you’re a Greek lawyer who has to explain what is going on to a foreign client, can you hand-on-heart say these various “English” renderings adequately convey what the Greek text says?
Would you feel comfortable relying on them in a professional context?
With tools like these you may gain speed.
“Translations” may be available at the click of a button.
But are the “translations” actually useable?
If you need a translation to understand a legal text or you’re a lawyer who needs one in their legal work, use a specialist legal translator.
Reach out to us at Jurtrans Translations for impeccably accurate Greek-English legal translations.
Today my timeline is filled with stories of the Greek Minister of Justice wanting to use AI-powered machine translation tools for interpreting in court settings and translating legal documents in various judicial proceedings. Settings where attention to detail should certainly matter.
The law is all about wielding words accurately (or deliberately wielding them in deliberately ambiguous ways). But let’s focus on the scenario where accuracy is what you’re after because you want / need to understand your rights and obligations in a legal setting; you want/need to make an important legal decision, etc.
In a post I did yesterday I stressed that the constancy of legal words is important; such constancy is important for a host of important reasons lawyers will immediately recognise, just some of which are legal certainty and the rule of law.
So I thought I’d run a little test connected to something I was working on.
I asked ChatGPT to translate a legal provision from Greek to English:
The text to be translated
This is the output:
ChatGPT’s output
Sounds quite plausible and convincing.
But let’s not forget that MT/AI systems used in translation are known for creating the “illusion of fluidity”[1].
They’re also known for a whole series of other problems (omitting bits of the text, changing negative obligations into positive ones, inconsistent use of terms, made-up words, etc.) but they are not relevant to us today.
What’s relevant today is this “illusion of fluidity”. On first reading the translation seems to be ok. It seems right.
On closer reading, especially if you compare it with the original, you start to spot “issues”.
Are you even able to compare it to the original to be able to identify any issues?
In this particular case, the article comes from a convention that already exists in English so we can easily determine what is right and what’s not. The relevant article reads as follows:
The “actual” legal provision
Admittedly, the original and “translation” are very similar.
There are differences you can easily spot:
“Entitled to benefit” vs. “Entitled to avail himself of”:
“Willful misconduct on its part” vs. “His wilful misconduct”:
“According to the law of the court having jurisdiction over the matter” vs. “In accordance with the law of the court or tribunal seized of the case”:
“Considered equivalent to willful misconduct” vs. “Considered as equivalent to wilful misconduct”:
“Omission” vs. “Default”
Some are probably not that important (you get the general idea whether worded in one way or the other). Are you only after a general idea though? Or as a lawyer/client do you want to precisely understand what the text is saying?
Others change the legal meaning utterly.
It does get the specialist term “wilful misconduct” right. Other online tools get it wrong (deliberate poor management / deliberate mismanagement / wilful mismanagement).
Although I didn’t ask for it, ChatGPT added “its” “view” about the output generated:
ChatGPT’s view of its output
Was it asked to provide simpler, more straightforward phrasing? We see a clear translational strategy here: opting for plainer language to make the text easier to read. Probably not a bad thing. Not the appropriate strategy though in this context.
It makes the bold assertion that both the original English text and its “output” are legally correct.
So I asked the obvious question:
ChatGPT generated this response:
If lawyers use precise wording “to avoid ambiguity” why does this system generate a “rough” translation?
So while I was focused on the details, and getting the translation “right”, our online tool was not.
ChatGPT provided an unofficial rendering of the provision. While the meaning may be “roughly the same”, it does not carry the same legal weight or and certainly doesn’t contain the precise wording as the official version.
Why does this matter?
Words matter in law … even translated words.
Established wording needs to be maintained.
Established wording is what lawyers will recognise and are used to working with.
Any deviation from established wording creates headaches.
Sticking to established wording saves users of the translation time, and avoids a lot of head-scratching and bewilderment of the type “so legally speaking what does that actually mean”?
Legal language is a “controlled” language. You cannot just use any old words you want.
Randomly generated “rough” translations introduce inefficiencies into lawyerly processes.
Randomly generated “rough” translations create false impressions of legal rules / legal obligations, especially if you aren’t an expert in that area of law; say, for example, you’re the client rather the lawyer.
Basically with tools like this you get “a translation”.
But …
Is it “the translation” you need?
Is it a translation you can use?
Is it a translation you can rely on in your lawyerly dealings?
Is it a translation you can trust?
It is a translation you can base decisions on?
Work with expert legal translators if you need your legal words to count.
By using this site, you agree that we may store and access cookies on your device. AcceptRead More
Privacy & Cookies Policy
Privacy Overview
This website uses cookies to improve your experience while you navigate through the website. Out of these, the cookies that are categorized as necessary are stored on your browser as they are essential for the working of basic functionalities of the website. We also use third-party cookies that help us analyze and understand how you use this website. These cookies will be stored in your browser only with your consent. You also have the option to opt-out of these cookies. But opting out of some of these cookies may affect your browsing experience.
Necessary cookies are absolutely essential for the website to function properly. This category only includes cookies that ensures basic functionalities and security features of the website. These cookies do not store any personal information.
Any cookies that may not be particularly necessary for the website to function and is used specifically to collect user personal data via analytics, ads, other embedded contents are termed as non-necessary cookies. It is mandatory to procure user consent prior to running these cookies on your website.