Me Michel Bergeron, partner at McCarthy Tétrault.

Enlarge

Me Michel Bergeron, partner at McCarthy Tétrault.

“You can’t beat the machine. Change is inevitable, there is nothing we can do about it. “

These words are from Me Michel bergeron, partner at McCarthy Tétrault. He explains how, in about two years, the firm has succeeded in making high-performance artificial intelligence for translations of legal documents.

“There are two phases, the first is the translation itself, then there is the revision. This is carried out by a reviser who must give an opinion on the translation and verify that the two documents, in both languages, mean exactly the same thing, ”he explains.

Today, McCarthy Tétrault employees are no longer called upon to do the raw translation stage.

But it took a long time for the machine to perform. “The result was not very good at the beginning because the translation varied too much from one area to another. Today, we have translations that take the context into account, because the word is translated according to the other words in the document, ”adds Me Bergeron.

If today the machine is more efficient, it is because it has been trained. He was given thousands and thousands of words to integrate. It’s a fact: the more database it has provided, the more it will be trained to do its job well.

Easier from French to English

Me Bergeron concedes that the translation remains more efficient from French to English, but it is more a question of language. “There are more words in the French language to describe the same thing. Conversely, translating a document from English to French is more complicated. The machine can skid, ”he says.

But at its best, the machine translates text almost in real time. Me Bergeron suggests that she will only take about fifteen minutes to translate a 500-page document.

Another example, a merger and acquisition project required the translation of 475 documents – totaling 2.8 million words – from German, Portuguese and Spanish into English in 10 days. . The result was that the firm’s AI reduced turnaround time and reduced costs to 45% of what they might have cost in the past.

“The best translator may achieve a better result, but remember that there is always a reviser who comes after the machine,” adds the partner.

In a course he recently gave to McGill law students, Me Bergeron pleaded for a change in teaching. Students must now be taught to spot the mistakes that AI could make, according to him.

The lawyer also underlines another advantage which is not negligible: the confidentiality of the data.

“We cannot afford to have confidential documents translated in Google translate for example. It goes to servers located in other countries, hackers can get their hands on it, even if Google says it stays in the cloud, it’s not as secure, ”explains Me Bergeron.

While with this AI, the firm’s data is stored in a secure server in Montreal.