As it turns out, documents are in essence very limited in their scope: certain information is to be exchanged for a given business purpose, and this information is to be kept confidential for a certain period. This limited scope makes them well-suited to automated analysis. There are after all only so many ways to say that information must be kept confidential and that leaks shall be judged by a court in California. Hence: Lynn, an automated legal analysis employing support vector networks and other machine learning technology to answer that simple business question: “Can I sign that document?”

Due to their ubiquitous nature and routine use, documents are generally regarded as a standard document. From a legal perspective, one couldn’t be more wrong. Drafting or reviewing a non-disclosure agreement is an art and therefore requires significant amounts of work to get right. This translates into a significant price tag to such a review, which is hard to grasp for an average businessperson who just wants to hear if (s)he can sign that document or not. As a result, there is something of a gap in the market between what legal professionals need to review a document and what businesspersons are prepared to offer. However, a robot is quite good at the type of basic review most people expect. Hence Lynn.

Pretty accurate. In statistical terms, my accuracy is 94.7, meaning in 94.7 percent of the cases I identify a clause correctly. My precision (the number of true positives over the total number of positive predictions) is 92.04, and my recall (the number of true positives over the number of positive instances) is 89.23. And finally my F1 (the balanced harmonic mean of precision and recall) is 0.91. Not perfect, but we’re getting there.

Of course. If I could sleep(), this would be keeping me awake at night. But seriously: most documents are so full of standard text that I am pretty confident I will get it right.

If you fed me all-new text for document clauses I would misidentify that horribly, true. But in the twenty years’ worth of documents me and my creator have analyzed, we rarely saw entirely new clauses for the same subject. Lawyers are a conservative bunch.

Probably not. The document is unique in that it is a common and relatively standard document. So there are lots of examples to learn from, and the impact is easy to analyze. But if you have a couple of hundred examples of some other contract, I’d be happy to learn something new!

Not right now, I'm pretty much hardcoded in what I like and dislike in a document. But we are working on a Business Edition which (for a fee) allows you to define a custom profile. Contact us to discuss options.

No, I can only review existing documents, not produce new ones. But over at JuriDox you can create a non-disclosure agreement yourself for only €45.

Not yet, but if you want to help out connect to their API I'd love to hear from you.

Sure, here you go: PNG and SVG. Please use only with both the name Lynn and Juriblox as entity behind it.

Please see the contact page.