As a developer it’s important to know your trademarks. But because of the arcane way that trademark law operates, making sure that your perfectly crafted company name hasn’t already been snatched up can be both costly and time-consuming. A Finnish startup called Onomatics, founded by four software engineers and trademark lawyers, thinks its new proprietary search engine is the answer.
"At least currently, most lawyers are using fairly traditional databases which only allow wildcard searches," says cofounder and IP lawyer Anna Ronkainen. "This is a massively inefficient way to operate, and it’s an area that computers can be used to do so much better."
With Onomatics, Anna and her colleagues have built the world’s most advanced trademark search engine, based around a unique artificial intelligence model of the trademark domain. As impressive as this is in legal terms, it also has relevance (and plenty of complexity) as a computer science project. Trademarks can take a variety of forms, ranging from words and sounds, to symbols and even colors. Trademarks don’t normally just apply to a name, but to a specific product and design (Coca Cola’s swirling calligraphy, for instance).
This equates to an enormously hard computing problem: classic search engines are less than ideal for the task, since what is required is a poly-dimensional search, not only equipped with the right legal datasets and algorithms, but additionally able to search along several lines of enquiry at once.
Onomatics’ TrademarkNow system is, Anna says, the only automated trademark system to search on these metrics. Having just closed its seed round with angel and VC investors from Silicon Valley, London and Scandinavia, Anna and her colleagues are hoping their creation will prompt a fundamental shift in the way that trademark analysis is done.
So how exactly does it work?
To perform a search with TrademarkNow, users are asked to input a trademark name (e.g., FastCoLabs), a product category (in this case a tech website) and the region in which you want the trademark to be registered (worldwide). From these three inputs, something called a similarity score is then generated showing how close your proposed trademark is to already existing ones. Similarity is described as a "safety level" between one and five, with each level accompanied by some broad legal advice on whether or not you should consider proceeding. (My proposed Fast Company knockoff, Speedy Business, was a recommended change.)
"Two fully identical marks are, of course, 100% similar," Anna says. "We define the 50% mark as the boundary point where the decision in a trademark opposition case should change. The big challenge is that currently trademark lawyers also express their similarity estimate as a percentage, and we have to try to match those figures to make sure our results ‘feel’ right, even if the percentages currently used have no objective foundation."
Rather than just giving a binary yes/no answer about whether or not a particular trademark violates laws or the rights of others, TrademarkNow compiles an in-depth report on the subject, listing all "similar" and "slightly similar" trademarks and the products those companies deal in.
It also fulfills another useful function. "The most important technical challenge for us is understanding the needs of the customer and trying to solve that," Anna continues. "For this reason, we could say we only deal with trademark law, but our customers don’t really have trademark law problems. The actual issues are naming and brand management, and while trademark law does play a key role, we are not afraid to add other data sources as well, such as dictionaries to get word meanings in hundreds of languages, and industry-specific data sources like names of mobile apps or pharmaceuticals."
Customers obviously don’t want to try and register a trademark that is already used elsewhere, but nor do they want to pick a name that has unforeseen implications. This is a mistake that even the largest companies have made. For instance, when Google first tried to launch its Chinese brand in 2006, it selected the soundalike name "GuGe" for its new service without realizing that the name means "harvesting song" in Cantonese and was therefore considered by many to be both strange-sounding and unsophisticated. By using natural language processing tools, TrademarkNow can draw upon hundreds of different dictionary data sets; potentially saving long-term embarrassment.
Of course, a bigger problem experienced by the team behind TrademarkNow is one that is common to those that have tried to combine law with computer modeling in the past: namely that laws don’t always lend themselves to automation. "The problem is trying to create a set of generalized rules matching the past decisions, because cases can be quite inconsistent and are often tied to the particulars of each individual case," Anna says. "Similarity of trademarks is tough, because there’s such a high level of subjectivity."
It is for this reason that the similarity scores TrademarkNow generates aren’t based solely on "objective" comparisons, but also on a prediction of how a trademark case might be decided were it to go to court. This is achieved using machine learning tools and a dataset of more than 50,000 previous case verdicts, which Anna describes as one of the largest case bases for intelligent legal technology in any field of law.
"Trying to come up with a reliable and comprehensive way to do this legal analysis computationally is basically an impossible task—especially when even the lawyers disagree on what some particular decision means," Anna continues. And while most legal cases only have a limited impact that could be incorporated into an existing system by, for example, tweaking the weighting of a particular algorithm, when it comes to precedent-setting trademark cases, decisions can sometimes have the effect of immediately rendering past decisions obsolete. For this reason, Anna explains that while the TrademarkNow system is automated, its updates are often made the old fashioned way: by having the team follow new cases and changes in legislation by reading the case studies themselves.
Anna started working on the project that became TrademarkNow around ten years ago, initially as a problem in computational legal theory. "I wanted to have a go at modeling laws using Type-2 fuzzy logic," she says. "Initially it was a matter of picking a domain pretty much at random. For a while I was considering doing this on Canadian tax law from the 1970s. Then I picked up a textbook on trademark law and realized that this was exactly the type of challenge I wanted to look at." After building an embryonic version of the TrademarkNow algorithm, Anna was approached by Mikael Kolehmainen, a trademark lawyer keen to enter the field of computer science, and the company was started soon after that.
The automation of law is one that is becoming increasingly important—seen everywhere from cars which have the ability to stop a drink-driver from operating them, to tech start-ups like Wevorce which promise to carry out divorce mediated by algorithm. There are two principle benefits to the automation of law: cost and speed. Although they’re still working on their pricing model for TrademarkNow, Anna points out that it is considerably cheaper than the traditional means of carrying out similar work. It is also faster.
"The traditional process can easily take a week to carry out the search," she continues. "Law firms will often have a lawyer representing the legal judgment, and a paralegal executing the search and dealing with the logistics. There’s a lot of back-and-forth. Our system takes between 10 and 15 seconds—no more than that."
What she now has to deal with is hoping that the rest of the system can catch up. "A lot of the trademark data storage is still very old fashioned," Anna says. "In some cases it’s available only on microfiche or index cards. Inputting it into a computer is something we will have to look at doing as we continue to expand."
There is no doubt that law is changing in the face of developments in computer science. But some parts are changing faster than others...