LexMachina: the translation machine for Swiss lawyers

Paula Reichenberg of Hieronymus AG about the ins and outs of the LexMachina website. She explains who the offer of the translation machine is directed at, which text types LexMachina can be used for – and why the job of a translator will stick around for quite a while.

Paula Reichenberg, you created LexMachina, the first neural translation machine that is specialized in Swiss law. What exactly is LexMachina, what are its capabilities and who is it for?

LexMachina is a neural translation machine that was trained specifically for Swiss law. It is a product by Hieronymus AG and allows Swiss lawyers and other persons and companies active in the legal or fiscal field to translate short and long texts from German into French, Italian and English as well as vice versa – with one mouse click. The great advantage of LexMachina is that it can deal with legal terminology and Swiss phrasing way better than conventional translation machines, as it was trained with Swiss legal texts.

Another plus is its security features – no data is saved during translation. All of the servers are located in Switzerland and the translation processes correspond to the security requirements according to ISO 27001 for information security. All of this makes it the probably safest translation machine that is directly made available to Swiss lawyers and legal experts.

How can LexMachina be used?

First of all it is important to know that LexMachina is standalone. This means that the operator can use the translation engine independently from other language services offered by the translation agency Hieronymus.

The translation machine is subscription-based: 75 Swiss francs per month allow the user unlimited access to the translation mask with a source and target field. You can upload and download whole documents (Word, Excel, PowerPoint, PDF, xml and others). The file format and layout of the original files are kept in the translated target files. As mentioned, thereby no data whatsoever is saved on the servers.

Additionally, the users receive access to their personal customer account. This allows them to commission additional services, such as post-editing of the machine translation by a specialized linguist.

So the entered data is not stored. This means, it is also not saved for the customer, right?

Correct. The data only remains on the LexMachina servers during the translation process – so only for a few seconds or minutes. Neither the source text nor the target text is stored, also not for future translations.

If the users need editing, proofreading or a notarization, they upload the document separately. In this case, the document remains on the server until the service has been provided.

Is LexMachina only available for lawyers?

Our solution was indeed first and foremost created for lawyers. However, it is also very much suitable for auditors, banks and insurance companies. We offer customized solutions for these companies – adapted LexMachina engines that align with their specific needs. Companies can use their tailor-made engines on their own servers.

How does LexMachina handle terminology and particularly customer-specific terminology?

We make sure that LexMachina uses the correct terminology with the help of following strategies:

1. Specific training with specialized data:
We train our translation engines and sub-engines with a large quantity of relevant Swiss specialist texts. Sub-engines are machines that are specialized for different legal fields or individual document types, for example criminal law or statutes. They can be selected accordingly before translation. LexMachina automatically uses the matching style and the fitting terminology due to the training with specialist data.

2. Dynamic Learning:
Additionally, the dynamic learning technology allows us to train the translation machine in a timely manner. The engine therefore learns from the corrections done by the translator in post-editing.

3. Tool for terminology management:
The desired terminology can be «forced» thanks to this tool. This approach is also called terminology integration. Thereby, the user uploads their own terminology that the machine ought to use.

How is LexMachina trained and how often?

In order to train a NMT machine (NMT = Neural Machine Translation) such as LexMachina, it is fed with a multitude of mono- or bilingual data. The deployed neural networks recognize various patterns and rules within this data and, from this, determine a single formula – a so-called algorithm. With this algorithm, the machine learns the connections between two languages.

This procedure, the training, can take several days or even weeks and takes a lot of computing power. A new training is only worthwhile if there is enough new data – around once or twice a year. However, the algorithm can continuously be adapted in-between the trainings thanks to dynamic learning.

What kind of texts are best translated with LexMachina?

Machine Translation is best done for texts that have recurrent sentence or paragraph patterns, deal with similar topics or are used for the same purpose – for example, Swiss employment contracts, statutes of a stock corporation or a limited liability company, general terms and conditions, share purchase agreement, listing prospectus, exchange offer, etc.

Hieronymus also uses LexMachina internally, when it is suitable. The translation processes remain the same in these cases, too. The text is post-edited by a legal translator and proof-read by a proofreader with a Swiss lawyer’s license. The whole project is supervised by the responsible project manager who also takes care of additional final proofreading before delivery to the client.

So the machine translation cannot completely replace the job of a translator?

Pure machine translation can suffice if a translation is only needed for general understanding, for example in order to facilitate daily communication between customers and colleagues. This is called purposeful Machine Translation. Texts that are bound to be published or have legally binding character, proofreading executed by a specialist or a professional translator is indispensable. Even a well-trained machine has its flaws, such as inconsistencies or “creative” translations of terms that are unknown to the translation engine. There is never a guarantee for error-free machine translation. Thus, post-editing done by a professional linguist or specialist is absolutely essential.

Thanks to LexMachina, the lawyer can do this task themselves efficiently – with due observance of his confidentiality obligations.

If an error-free translation needs additional post-editing: is this offer really still attractive, pricewise?

Even if you include post-editing, LexMachina remains the most cost-effective and time-saving solution for Swiss lawyers and legal experts. The customer does not only save time and money thanks to machine translation, but at the same time does not have to waive legal precision or confidentiality. We are very proud of this.

Paula Reichenberg, thank you for this interview.