As lawyer–linguists and seasoned legal translators we support fellow colleagues with the translation of legal terms in cases in which dictionaries and the internet do not help (see the details here), motto …
with long-standing experience as legal translators.
Whereas in our capacity as legal language expert and legal translation specialist, in addition to the aforementioned collegiality assistance that we offer on a not-for-profit basis, we provide on a commercial basis:
in particular, English contract translations, Dutch contract translations and German contract translations, moreover English-language, Dutch-language and German-language General Terms and Conditions, by experienced specialist translators in the field of contract law, rather, the law of obligations, and commercial law.
Translation? legal translation? Does it not seem a bit outdated since machine translation providers offer the translation of entire legal documents with just one click?
In this regard, it would be advisable to consult the general terms and conditions of these providers and, in particular, the provisions on liability and warranty before engaging in any such endeavour!
However, with the advent of artificial neural networks, machine processing of natural language, neural machine translation and large language models (see below) we as lawyer-linguists consider ourselves as professional in:
Furthermore, we are a supplier of linguistic professional support in the field of interstate (public) law, especially in the translation of a data protection declaration in the course of the GDPR.
by experienced lawyer-linguists in the capacity as a specialist in the translation of legal texts of any kinds, expert in legal terminology in general and authority in multilingualism in law in particular.
What exactly is this all about?
It is about reviewing a legal translation on a high end level. And, in doing so, investigating (examining) for:
(a) inconsistencies to clarify and correct,
(b) contradictions to recognize and eliminate,
(c) common misconceptions to identify and rectify,
(d) common errors to discover and resolve,
(e) obvious deficiencies to spot and obliterate,
(f) hidden omissions to find and supplement,
(g) possible misunderstandings to uncover and clear up,
(h) unspecified wording to reveal as such and avoid,
in order to guarantee the authenticity and integrity of a legal translation.
in the broadest sense, that is machine translation of specialized legal texts by way of
And, what exactly is that all about?
It is about the use of what is called a foundation model, again in the broadest sense.
In the broadest sense of what?
In the broadest sense of what is meant by steering such a foundation model or any type of language model toward the most desirable outputs in the field of law in general and in the field of legal translation in particular.
Albeit, it means that the role of an authority in the field of automated legal translation is threefold.
That is, look through the eye – and thus combine the role – of:
(a) a common linguist in the capacity of a reviser,
(b) a specialist linguistic reviewer in the capacity of a technical editor with basic understanding of computational linguistics and, foremost,
(c) an expert in the legal system of the source language and of the target language in the capacity of a legal editor,
What does this mean?
In the course of a sound and sophisticated post–editing, as described further below, one has to, rather needs to skilfully cover up and compensate shortcomings and correct errors in machine-translated sentences in order to ensure coherence and consistency in translation.
Such a review shall take place on the basis of considerable familiarity with regard to the automated translation process, implementing certain state-of-the-art quality metrics, specifications and requirements.
And, what does this imply?
Familiarity with artificial intelligence in general and machine translation in particular
What exactly does such familiarity refer to?
Basically, it refers to five different things or rather concepts to be familiar with:
(a) Natural Language Processing (NLP)
(b) Large Language Models (LLMs), also known as foundation models, like, among others: OpenAI’s ChatGPT and GPT-4 LLM, Alphabet/Google Bard and LaMDA and Pathways, PaLM 2 and Gemini, Anthrop\c Claude 3, Meta LLaMA
(c) Neural Machine Translation (NMT) Deep Neural Networks, provided by, among others: DeepL, Tilde, Microsoft Translator European Union/e-translation and ModernMT
and, of course, as a key concept inextricably linked to NLP, LLMs and NMT:
(d) Deep Learning (DL) in the context of Machine Learning (ML), and
(e) what is called Reinforcement Learning from Human Feedback (RLHF) as a mean for direct interacting between artificial intelligence and, if you like, natural intelligence, along with what is called Transfer Learning (TL).
To illustrate this, that is to demonstrate basic knowledge of computational linguistics procedures in relation to legal translation, there are four steps to be taken:
.And further, what does this imply in detail?
Abstractly, it is about the procedures of the so called pre–editing machine translation and post–editing machine translation.
Concretely, it means the gradual development in becoming a specialist, even an expert in:
(a) reviewing and adapting legal texts for automatic translation into a foreign language (pre-editing for that matter, namely editing ex ante). Part of the process and of great importance is the anonymization of such text, thus the removal of any personal data that is any information related to an identified or identifiable natural person.
In other words: implementing natural intelligence as a prior step vis à vis the implementation of artificial intelligence.
Apart from the need to remove personal data such a pre-editing of text is, needless to mention, not always necessary, but essentially so if text:
– has errors and thus mistakes in orthography, grammar and syntax or has no punctuation or wrong punctuation – because only linguistically error free and thus accurate text can be automatically translated with maximum precision,
– is not objectionable in its actual wording but consists of sentences that are difficult, clumsy, confusing or incomprehensible and thus lacks linguistic accuracy – because only linguistically consistent text can be automatically translated with maximum precision,
– lacks a logical overall structure or shows a discontinuous structure or an awkward overall structure – because only linguistically coherent text can be automatically translated with maximum precision!
(b) correcting, which means technically and materially editing automatically translated foreign-language legal texts (post-editing for that matter, namely editing ex post) on the basis of and in accordance with the parameters, as set out further above in the legal proofreading and legal editing – review and revision section.
In other words: implementing natural intelligence as a follow-up step vis à vis the implementation of artificial intelligence.
Apart from the need to apply the criteria mentioned above in section review and revision there are additional points which one has to keep in mind, for example, words or phrases that appear in the translation but not so in the source text or, the other way round, words or phrases that appear in the source text but not so in the translation, or a sequence of words that has slightly become out of order, and although not immediately recognisable as such is causing a most awkward misinterpretation of meaning: be aware that this latter matter is one of a very tricky nature with any machine translation endeavour!
How could all this be brought together in thought and in word?
Artificial intelligence searches for the right wording, natural intelligence searches for the right words!
