Making the right terminological decisions in EU translation is a very complex and demanding cognitive process. It requires extensive research, comparative skills, understanding of context and the given domain, and also in case of lawyer-linguists sometimes comparative law analysis.
The art of making terminology choices
At the Translating Europe Forum (#2025TEF) which has just been organized in Brussels last week under the main topic “Quality matters”, it has been emphasized that, on the one hand,
- quality has to be preserved, and on the other hand,
- after applying a thorough risk assessment, the level of quality required depends on the purpose of the text and the intended audience.
Nevertheless, it was universally agreed that in case of high-risk texts, like legal, financial, medical translations, quality means adequacy, accuracy and strict compliance with specifications.
In her opening speech, Mrs. Valeria Darò, Director of DG Translation of the European Commission said that the DGT creates translation memories using previous translations,
“so that not a single sentence has to be translated again if it is really the same”.
What we built at Juremy serves the same purpose: we provide easily accessible and searchable aligned segments on the curated EU corpus, so that linguists can quickly find previously translated phrases to enhance quality of EU translation.
Shortcutting the process of finding previously translated equivalents is a significant productivity improvement in itself.
BUT after being presented several bilingual results, linguists enter the next phase of the expert process: choosing the right terms.
This step needs professional expertise, knowledge, experience and plenty of … TYPING THINKING.
How to decide which term to use?
⬇️ In our example below, the same French term (“objets mobiliers corporels”) has two totally different Hungarian translations found by Juremy.
⚖️ As the screenshot shows, the EUR-Lex reference documents are also different: the earlier is the Brussels I regulation (44/2001), and the more recent one is the Brussels Ia regulation from 2012, which has recast the former document.
To decide which wording should be used in Hungarian, the date of the relevant facts of the case should also be taken into consideration:
▶️ If the Brussels I regulation is applicable to our case, the corresponding terminology used in its Hungarian version (“ingó dolgok”) should apply accordingly - after checking that no corrigendum has affected this specific wording.
⏩ However, if the facts of the case are regulated by the Brussels Ia. regulation (1215/2012), the term “áruk” should be used to translate the same French term.
On the screenshot above, you can see the blue mark next to the hit from the Brussels I regulation - it shows that I have added this document as a preferred document, using the “Reference documents” feature. The reason is that in my source text, this earlier version of the regulation shall be applied.
💡 Not so straightforward, is it? This also shows that it requires careful consideration which result to use.
That’s why in our opinion it is very important to be able to see multiple terminology choices from reference texts, instead of displaying a single ready-made answer to a query - the latter can entail significant risks.
The risk of using NMT suggestions for already translated sentences in the EU corpus
As it was emphasized in the opening speech at the TEF, it is crucial that if a sentence has already been translated and the context is also the same, this previous translation should be preserved to ensure consistency among EU texts.
That’s why there is a risk of using NMT-generated texts to provide translations for previously translated segments or sentences - or even terms - in EU texts.
What are these hidden references and why is it important to find them quickly?
As mentioned in our previous blog article, EU texts - e.g. opinions of the advocate general to mention the legal domain - contain so-called hidden references.
These are citations or references to previous texts which are not necessarily marked by footnotes, but which refer to prior case-law or legislation, sometimes in a rephrased wording.
It is problematic when citations or hidden references, which have already been translated in EU legislation or case-law, are being re-translated by NMT engines.
Why?
Because if identical source text sentences are reproduced in different target language variations, it causes inaccuracy, inconsistency and legal uncertainty.
The example below shows an official EU legal text - a judgment - which has an English and French aligned version in EUR-Lex. We found the exact match with Juremy in a second.
After that, we generated an NMT translation of the same source text with a commercially available NMT engine.
On the comparison chart below, you can see the terms and collocations highlighted where the NMT tool translated the original English term in a different way (in red) than used in the EU corpus (in green). Naturally certain degree of variation is permissible, but in case of established legal terms the usage of different target language versions can lead to inconsistency and inaccuracy.
How can these risks be avoided during the EU translation workflow?
- First of all, the linguist has to be able to detect these hidden references in an efficient and fast way.
- After identifying the source of these references, linguists have to be able to quickly find the target-language equivalent of these phrases or terms.
- If these tasks are difficult to fulfill using the existing resources available to linguists working for EU institutions, there is a risk that already existing matches of official EU-precedents are being re-generated and re-translated in a different way, resulting in inconsistency of EU texts in the longer term.
Instead of using NMT suggestions for already translated sentences, find the equivalents with Juremy in a second
It is clear that, to ensure consistency and high quality in translation, using these previous translations is essential. Otherwise, there is a risk that already existing equivalents are translated in different versions, resulting in inconsistencies and legal uncertainty throughout the EU legal corpus.
However, it is very difficult to find these references in the entire EUR-Lex corpus (with more than 320.000 documents per language) without specialized search functionality offered by Juremy.
Without Juremy, linguists have to browse through an indefinite number of bilingual documents searching for the previously translated sentence, which is a time-consuming process and not always fruitful either.
Juremy makes it possible to see a targeted list of these previously translated segments or phrases in a second, saving the linguist a huge amount of time on each task. Running Juremy through an entire document via the CAT-tool plugins makes it even faster to immediately find the concordance matches for each sentence in the source text.
This way, it can efficiently be verified if the post-editing of an AI-suggestion is superfluous, since the translation or its fuzzy match already exists in the EU corpus.