Many companies today are employing deep learning techniques in different facets of their business. Yelp uses deep learning algorithms to feature the best user photos, Netflix uses it to suggest movies you might be interested in, and Google ultimately transformed the concept of deep learning by creating a system that helps generate responses to search queries.
It’s widely believed that you no longer need to know your data; you can just apply a little deep learning magic and poof — problem solved. However, the reality is that this could not be further from the truth, at least for the legal industry.
As an example, deep learning can be essential when legal counsel within an organization wants to find out how many contracts (among 10s to 100s of thousands) has termination for convenience clauses that could disrupt the business, or if any have strict assignment rules that may be a problem for a M&A event. This process would normally consist of manually reading every document, but with deep learning, that process can be automated. The same holds true across other parts of the business as well. With procurement needing to find how many auto renewal clauses may be hidden within vendor contracts, or sales needing to know how many customers have 45-day payment terms vs. the standard 30, these are the needle-in-the-haystack scenarios that deep learning can automate, dramatically reducing time and cost.
Deep learning networks have significant capacity and promise. This great capacity implies that you need to design your networks carefully to be able to find the structure in your data, which in turn requires domain knowledge. You can’t use an out-of-the-box approach and expect world-class results. Instead, you need knowledge to properly adapt your plan and tools directly to your specific task. In order to succeed when applying deep learning methods to legal domain data, you need to truly understand that data.
Languages are neither uniform nor standardized — they have quirks. Semantics are often hidden in metaphors, idioms, and colloquial expressions, not to mention the explicit obfuscation of meaning by complex embedded structures. Because of this, deep learning methods for natural language processing (NLP) aren’t as simple as something like image recognition for visual data.
In today’s visual domain — think Facebook, Instagram, Snapchat, and others — data is everywhere. However, when it comes to the legal space, the volume of data is harder to come by. And it’s pricier. This lack of data makes it much more important to understand and correctly handle it. As far as deep learning goes, this means software needs to construct networks that are able to capture the meaning and structure of legal terms, clauses, and documents.
Unfortunately, the size of the data with examples of legal provisions can be too small to train deep learning networks; furthermore, a specific provision consisting of only one sentence may or may not exist in a thousand-page contract. So what can legal teams do to successfully find a provision consisting of only one sentence that may or may not exist in a thousand-page contract — the equivalent of locating a needle in a haystack?
With today’s advancements in computational power and technologies like machine learning, there are a plethora of networks and models available to the masses. From Google’s Machine Translation to my company Seal’s deep learning branch of machine learning, companies are finding ways to deliver products that are ready to be applied to customers’ data. In the legal industry, this will help us to achieve the desired result: getting the job done more efficiently.
Kevin Gidney is the cofounder and CTO of Seal Software, a contract discovery and analytics software.