Machine translation, artificial intelligence and the limits of legal language
The promise of machine translation
Since the late 2010s, neural machine translation has transformed the practice of translation. Systems such as DeepL, developed in Cologne, and the translation functions of large language models produce translations between German, French, Italian and English that are often fluent and in many cases accurate. For the Zurich financial centre the new tools promise to lower the transaction costs of multilingualism dramatically. A German-language judgment of the Federal Supreme Court can be rendered in English in seconds. A German-speaking client can read a French-language contract without waiting for a human translator, while an English-language board paper can be summarised in German for a Swiss board member. The economic theory outlined at the beginning of this chapter predicts that such a fall in transaction costs will increase the volume of cross-language transactions and reduce the premium that individuals earn for their language skills.
Public administrations have taken up the new tools as well, using machine translation as an aid for their translators. Courts and law firms in Zurich use machine translation for the rapid review of foreign-language documents, particularly in disclosure-heavy disputes and in internal investigations. Chapter J.III has discussed the professional secrecy constraints that apply when lawyers and banks feed client documents into cloud-based systems. Those constraints apply to translation tools as well.
The limits for legal language
The limits of machine translation become apparent precisely where the linguistic infrastructure of a financial centre is most demanding: in legal terminology. A system trained predominantly on German-German texts will tend to translate the Swiss Verwaltungsrat as “supervisory board”, the Betreibung as “enforcement” and the Revisionsstelle as “audit department”, each time importing a German or English concept that does not match the Swiss institution. It may render Konkurs as “insolvency” and miss the distinction between enforcement by seizure and enforcement by bankruptcy. It will rarely recognise a Baurecht in a Swiss land register as a property right. Terminology databases such as TERMDAT, with its multilingual entries on federal legal terminology, can mitigate these errors if they are integrated into the translation process, although the underlying problem remains.1 Because legal terms name concepts within a legal system, a translation system that does not know that system cannot choose the right name.
A second limit lies in authenticity. A machine translation of a Swiss statute into English has no legal force, just like the official English translations on Fedlex, which carry the notice that English is not an official language of the Confederation.2 The binding text remains the German, French or Italian version, whose interpretation requires knowledge of Swiss doctrine and case law. A third limit concerns liability. Whoever relies on a translation in a legal transaction bears the risk of its errors. Human legal translators and lawyers remain responsible for the accuracy of their work, while the providers of machine translation disclaim such responsibility. For contracts, prospectuses and submissions to courts and authorities, the professional review of machine translations by qualified lawyers or legal translators therefore remains indispensable.
Consequences for the financial centre
Machine translation will change the economics of language in the financial centre without abolishing them. It lowers the cost of access to foreign-language texts and makes routine communication across languages easier. It shifts the value of human language skills from producing translations to checking them and from general fluency to specialised terminological and legal competence. The Zurich lawyer who knows the difference between the Swiss and German meanings of Baurecht, between the Swiss Verwaltungsrat and the German Aufsichtsrat, and between Konkurs and Insolvenz, becomes more valuable, because these are the points on which machine translation fails and on which errors are expensive. The same applies to the relationship dimension of language. No machine can speak Züritüütsch with a client in a way that builds trust. Nor can a translation system replace the shared references that make a dialect a carrier of social capital.
The outlook is therefore one of continuity in the layers and change in the tools. Standard German will remain the language of Swiss law and of the German-speaking hinterland. The dialect will remain the language of personal relationships among the Swiss. English will remain the working language of international finance and will gain ground in courts and supervision, always as an option added to the official languages. Machine translation will lower the barriers between these layers, while the specialised knowledge of what lies behind the words will become the scarce resource of the linguistic infrastructure.
Notes
- Schweizerische Bundeskanzlei, “TERMDAT”, loc. cit. ↩
- Conflict of Laws.net, “Out now: the Swiss IPRG in English”, loc. cit. ↩
For general information only. This article does not constitute legal or tax advice, does not replace advice on an individual case and creates no lawyer-client relationship. It reflects the legal position as of September 2025.