Recently, I had a chance to talk to the Sean Martin, CTO of Cambridge Semantics, a Massachusetts-based company that has developed a semantic middleware stack, complete with ESB. We largely discussed how companies are applying the company's solution to enterprise data that's thus-far un-managed and un-integrated: spreadsheet data.
Earlier this month, IT Business Edge ran my Q&A with Martin, but I didn't include in the published interview our discussion about SemTech 2010, which Martin attended. I asked him about any integration-related announcements at the conference.
I was surprised to learn that, while semantic vendors don't always talk about it, Martin says integration is central to any semantic solution:
They may not have our Excel capabilities and so on, but they have a story around integration of data. That's really the first thing you've got to do to start the benefit from a semantic stack is you get the integration layer down. And then after that, you can start to think about inference and reasoning and more sort of intelligent applications. But the first layer is the integration. So you'll find everybody, all the people who build semantic companies, that will be one of the prime things that they will do. It's the first thing they'll do when they engage with you.
You have to wonder, then, why this integration angle is so seldom discussed when you read about semantics, particularly when you start to move into applying semantic solutions to the enterprise. Martin has a simple theory about that: Semantics comes from academics, and academics tend to be more interested in "esoteric things that you can do," he says. So while you can't even begin without a solid integration foundation, that's just not what interests them about semantic approaches. Academics and many of the semantic startups are more focused on the applications they can build, not on how semantics affects that integration layer.
It's not hard to see how semantics can affect enterprise integration, though, when you look at how Cambridge Semantics is applying it:
Most companies are actually thinking about their semantic stacks in more of a siloed-type solution. They think about it as being able to integrate data, but what they're actually building are some application or other, and that's a huge contrast to how we think about it here.
We have this notion of this semantic fabric into which you map all of your data sources over time. As you map each source, that's effectively making it available via a SPARQL-type interface providing an ontology to describe each interface, each data source, (and) putting those into directories. Then to do integration becomes much easier, because two-thirds of the job has been done and you've got a reusable information asset ready to go. The end users can often do a lot of integration themselves, we have tools that allow end users to do visual integration.
It sounds very promising, and certainly we're beginning to see more about how semantics might be applied to more "traditional" enterprise problems. For instance, this week, researchers at the Sheffield Hallam University in the UK received 370,000 (US$469,126) to develop a platform that combines semantic technologies with business intelligence techniques to obtain online and offline data, then visually analyze it.