Cortical, or Cortical.io as it likes to be known, started 10 years ago as an academic research project and has evolved into a contract analysis tool founded on semantic folding, an approach that seeks to give contextual results based on overall meaning.
In this AL TV Product Walk Through, CMO Steve Levine, takes us through some of its capabilities, especially the way that it takes a semantic approach to contract analysis and extraction. For example, it can extract a date from a lease and assign it to the right legal term even if the clause where the date is located doesn’t mention that term. It does this by looking at the context.
It can also compare two clauses from different versions of the same document and show if they are similar, based not on the exact word use, but the actual meaning of the words used as a whole. I.e. two clauses may appear to be different on the page, but Cortical will analyse them and show they have the same meaning.
Levine also stated that Cortical can get reliable results with a training set as small as 50 to 100 documents. This again is largely down to the semantic folding approach, which he described as a different methodology to the statistically based approaches used by many NLP systems.
The video is about 20 mins, and the initial walk through covers the first 11 minutes, the remainder is Q&A. Press play to watch inside the page.
This site found it quite impressive, all the more so because Cortical has not made a major effort to work with lawyers, in part because they’d been told in the past that lawyers were not that interested. Instead, they have focused on other sectors such as management consulting. Clearly there is an opportunity now to return to the legal world, both to law firms and inhouse teams.
But, back to the semantic folding. There is a paper here that explains what it means in more detail. Or check out the Wikipedia page here. The basic gist is that they take a ‘representational’ approach to meaning that is derived from neuroscience studies of how the brain understands information and ideas.
For now it primarily works in English and also German. If you fancy giving it a go this site would be very interested to hear what results you get.