Screen Shot 2013-10-02 at 10.23.52 PM

With an increasing number of commercial applications relying on its technology, natural language processing (NLP) has started to come into its own of late. After years of research and experimentation, it has emerged from geeky obscurity and into the limelight — thanks to Watson, Siri and Nuance’s Dragon, among others. Of course, every time Siri misunderstands a command, it’s easy to dismiss those aforementioned decades of progress, curse technology and chalk it all up to hype.

Of course, doing so is dismissing the fact that NLP is tackling a near-impossible task: Teaching computers to understand human speech. Deconstructing and understanding the endless string of words, expressions and sentences, and the nuance of syntax, emphasis and tone in human speech is a byzantine problem. Even for super computers. Part of the reason for this, says SkyPhrase founder Nick Cassimatis, is that those brave enough to tackle the problem in the first place have generally stuck to two main approaches: Using statistics and brute force or by way of logical reasoning and language rules.

Peter Thiel’s Breakout Labs, a new investment vehicle that backs startups pursuing cutting-edge research and hacking away on problems that “might still be too risky to get traditional venture backing,” recently selected Skyphrase as one of its first investments. NLP as a whole is still in its infancy and much of the hardcore innovation and engineering is taking place within academia, making it a tough (and risky) bet for investors.

However, while SkyPhrase’s founding team is composed of researchers and professors itself, Cassimatis wants to not only bring his work outside of academia but compete with the Siris of the world by taking a third, less-traveled and trickier approach to the NLP problem.

Skyphrase, as Kim-Mai wrote in April, has developed an artificial intelligence technology that aims to combine both the brute force and logical reasoning methods mentioned above, and take a more holistic approach to teaching machines how to understand human speech. Led by CEO Nick Cassimatis and a handful of his graduate students from Rensselaer Polytechnic Institute (RPI) outside of Albany, the team collectively has decades of experience working in artificial intelligence and computational linguistics research at RPI, Stanford and MIT.

Over time, the team has developed algorithms that essentially work to synthesize logical reasoning and traditional processing in order to analyze a larger and more complex range of queries, ideally with greater precision. While it’s tricky to sum up nine years of research without being a little reductionistic, Cassimatis explains SkyPhrase’s alternative technique as one that focuses on teaching machines to respond to queries by helping them to analyze and build meanings piece-by-piece and brick-by-brick from available signals.

In contrast, tradtional NLP technologies may receive a user’s query and attempt to fit it into a pattern they’ve already predicted, he says, so, when you throw Siri a query she hasn’t heard before, for example, you’re probably not going to like the result. SkyPhrase is able to more accurately answer these types of queries, Cassimatis says, by leveraging both algorithms and data structures to paint a more holistic view of the components in language.

Doing so has allowed Skyphrase to produce results that are significantly more accurate than what one has come to expect from Wolfram Alpha and Siri, he says. In fact, the CEO claims that SkyPhrase’s NLP tech can respond to whole questions at close to 90 percent accuracy on average, where as Wolfram is closer to 30 percent.

While that’s encouraging, it’s important to remember that NLP and SkyPhrase itself are both in the early stages, and there’s still a pretty steep learning curve for the average user setting out to use NLP tech. The MIT Technology Review, for example, found that the startup’s tools still needed some training when it came to applying it to more unfamiliar territory, like in Twitter and Orbitz searches, for example. The artificial intelligence technology still requires some activity, the CEO explains, and it has to learn as it goes. In areas it was more familiar with, SkyPhrase can add a lot of value quickly, which is why the startup has decided to apply its tech to two new areas — in two distinct areas.

With SkyPhrase, Cassimatis really has two goals: One is to hone the NLP technology and continue to productize it in areas where it can add real value and, two, reduce the amount of legwork, friction and time it takes to develop natural language interfaces. “Our goal, long-term,” the founder says, “is to let as much of the world’s data be accessible using natural language as possible by building out applications in individual verticals, but also by making it easy for third parties to create natural language interfaces for their own data and applications.”

To start, SkyPhrase has developed two applications of its own to test and improve its technology, one being a search and analytics product which anyone can apply to their own Google Analytics account and the second being an app targeting fantasy football fans. Its fantasy football search tool allows users to quickly search statistics without being distracted by irrelevant links and is great for drafting and selecting the best players, making trades and so on.

The other cool feature of the app is that it allows users to set up alerts during the season, which could range from the simple, like “tell me when Tom Brady throws two touchdown passes today,” to more complex queries. This is also a potential area of differentiation for the startup.

While there are other cool apps out there like Chadwick and tools to help fantasy statisticians like numberFire (used by ESPN to power some of its predictions), others put most of their focus on past events and data, making the results a little less actionable than most fantasy fans would like.

In turn, its Google Analytics application helps users answer complex and precise questions about their web traffic, for example, without needing to learn and use a more complicated interface or rely on special analytics reports. Users can ask questions in simple language and get answers quickly, rather than having to trawl Google for answers. To give an example, SkyPhrase for Google Analytics can handle commands like, “show me pages that visitors from New York City viewed in the last three days,” or questions like, “which keywords generated the most traffic to our registration page last week?”

In the end, Cassamitis says, “data is full of important insights that are never discovered because the answers are too difficult and expensive to find.” While SkyPhrase is beginning with football and Google Analytics (because that’s where they wanted help), the CEO thinks the startup’s tech can make it easy for developers with “little knowledge of linguistics or artificial intelligence to create natural language interfaces for their own data and applications.”

It’s still early in the game, but it’s a mission that’s pretty easy to get behind, and each step the startup takes on the tech side could result in some pretty cool consumer and enterprise apps on the other.

For more, find SkyPhrase at home here.

Resources
Post Your Resume to 65+ Job Sites
Resume Service

Post to Twitter Tweet This Post