LexMachina. The translation engine that knows the law it translates
A neural machine translation engine trained on Swiss legal and financial data: over 15 million multilingual segments from Federal Supreme Court decisions, FINMA publications and the federal codes.
Request a demoThe difference lies in the official term
A financial text, translated from English into French. Look at the Swiss legal terms, not the style.
The company must deduct the 35% Swiss withholding tax from the dividend and pay it to the Federal Tax Administration. The capital increase is subject to the issuance stamp duty and must be entered in the commercial register. As a supervised institution, the bank also reports the transaction to FINMA.
Fast Lookup| What to look at | LexMachina | General-purpose AI |
|---|---|---|
| The tax on the dividend | impôt anticipéofficialthe name of the Swiss tax in the Federal Act on Withholding Tax | retenue à la sourcegeneric term |
| The bank's status | assujettiofficialthe term in Art. 3 FINMASA (supervised persons) | établissement soumis à surveillancedescriptive paraphrase |
LexMachina is trained on Swiss texts. It uses the authoritative term, where a general-purpose engine only paraphrases it.
Compare the engines on your own texts in the demoIn this example
- Actual outputs from both engines, reproduced word for word, recorded on 16 October 2026.
- Impôt anticipé (withholding tax) and assujetti (supervised persons): terms from the Federal Act on Withholding Tax and Art. 3 FINMASA, checked in CHnell.
- An engine's output varies from one generation to the next: judge it on your own texts, not on a selected example.
What LexMachina is, exactly
A specialised translation engine, developed in Switzerland, available in Corrext on the same footing as the engines on the market.
- 15M+training segments
BGE/ATF, FINMA, codes and Federal Acts. Never client documents.
- 4native languages
German, French, Italian and English, in every language pair.
- CHrun in Switzerland
Even outside Highly sensitive content mode.
- 4alternatives per sentence
Each one labelled with its engine.
- Areas of lawCO · contractsBankA · bankingFINMAPILA · arbitrationCompany lawTaxProcedure
- In Corrext, chosen for each requestText translationFile translationTranslation projects
- The other engines remain availableDeepL ProAzure OpenAI GPTClaude
Why a specialised engine rather than a general-purpose one
Not because it would be ‘better’ in absolute terms, but because Swiss law has a rare feature: it is already published in four languages, and that material can be learned.
The Confederation publishes its laws in German, French and Italian, and a large share of its texts in English. An engine trained on these texts does not look for a plausible equivalent: it uses the term of the official text. ‘Verzugszins’ becomes ‘intérêt moratoire’ in French, the term of the federal acts, and ‘default interest’ in their English translation.
‘Art. 104 al. 1 CO’ in French is written ‘Art. 104 Abs. 1 OR’ in German and ‘Art. 104 para. 1 CO’ in English. LDIP becomes IPRG, then PILA. An engine that has seen millions of these correspondences returns them in their official form instead of translating the abbreviation word for word.
In a 60-page contract, a concept must keep the same name from the first page to the last. That is the job of a vocabulary learned from consistent texts. It is also the job of the Alternatives panel, which lets you settle on a wording and then stick to it.
Further reading
Try the engine on real statutory provisions and compare the alternatives in the demo.
See the tool Security and data sovereigntyWhere your document goes, depending on the mode and engine you choose, stated precisely.
See the page Fast lookup CHnellThe Swiss law concordancer: official sources, 30 domains, your own resources.
See the toolFrequently asked questions about LexMachina
What is LexMachina trained on?
On Swiss legal and financial data: over 15 million multilingual segments from Federal Supreme Court decisions (BGE/ATF), FINMA publications, and the federal codes and acts. Never on our clients' documents.
Which languages does LexMachina cover?
German, French, Italian and English, natively and in every language pair: the four languages in which lawyers in Switzerland write and argue. The Corrext platform offers 30 languages in total. Beyond these four, a third-party engine handles the translation.
Where does the engine run?
In Switzerland, on Neur.on's Swiss infrastructure. Choosing LexMachina is enough to keep processing in Switzerland, even without Highly sensitive content mode. The Attorney-Client privilege notice does not appear, because nothing leaves the country.
Are my documents used to train the engine?
No. LexMachina learns from public Swiss legal and financial texts, not from your files. In Standard mode, your projects feed a translation memory that only your organisation can use. In Highly sensitive content mode, no translation memory is fed.
Can I use engines other than LexMachina?
Yes. Corrext is model-agnostic: DeepL Pro, Azure OpenAI GPT and Claude are available in the same interface, and you choose the engine for each request. Engines hosted abroad trigger the Attorney-Client privilege notice, which reminds users of your organisation's policy.
How do I compare engines on my own texts?
In Text and document translation, the spark button opens the Alternatives panel: up to four suggestions for the same sentence, each labelled with its engine. Generate more gives you further options. You judge on your own texts rather than on a promise.