|
|
BotSpot® Bi-Weekly Newsletter March 28, 2001 A Question of Semantics There are various levels of perception and understanding humans have when we look out upon the world. That's a key part of what makes us intelligent--the ability to look at a single thing and assign different labels to it based on the context of that thing. We can look at a single drop of water falling from the sky and say to ourselves, "this is water." Or, "look at the lovely raindrop." If the context of this drop of water includes its accompaniment by several million similar objects, our likely self-response will be "get in out of the rain!" It all depends on how you look at things and how things are currently positioned in the universe at large. Taking away all arguments of natural perception for a moment, let's simply focus on the written word. To call literary perception simple is a bit of a stretch, but I'm trying to avoid far-reaching philosophical arguments this week. When we look at a sentence: "His heart began to pound...", what do we take away from this? Some man's heart is beating fast. Almost immediately, we start to ascribe motives to this action. Is he having a heart attack? Is he in fear? Perhaps the rest of the sentence will help us. "His heart began to pound when she entered the room." Now this is a better clue. It's about a woman, something that's surefire to start a man's heart racing. This is probably about romance, then. Or does this man fear this woman? Or hate her? For that, of course, we would need the rest of the story. That would give us the context we need to figure out what's going on. The point here, of course, is that we automatically try to ascribe meaning to almost any sentence we read. We read beyond the words. This is something that very few products of technology can do. Bots, particularly search bots, do not do well with context. A bot would likely look at this sentence and pull out the words "heart" and "pound" as keywords, and then perhaps use them later as results for a search for meat prices. Bots may make some contextual assignments to data they come across, but usually this contextual judgment is simple and one-dimensional. Teaching context to bots is not an easy thing to do. One Massachusetts company is working to change this problem. Jarg Corporation is working on developing agent-based technology that goes beyond simple word and phase matching and into real context matching. With data piling up to our metaphorical eyeballs, getting the real nuggets of information out all of the useless data has become an industry in and of itself. What Jarg is doing is creating industry-specific search engines that use that industry's jargon and lexicon to form the basis of contextual relationships in the words it finds in stored documents. Jarg, therefore, may provide a medical research firm with a search engine that already understands key terms the firm may use. This immediately gives it a leg up on regular search engines. The next methodology it applies to searches is a process called Keynetting, which looks at the structure of the query and creates a contextual understanding of that sentence. In the example given on the Jarg Web site, the query "What are the emotional side effects of the treatment of colon cancer?" is compared with Jarg's Knowledge Engine and other popular search technology, such as Google and Infoseek. Whereas the latter types of search tools might break the query down into keyword components, Jarg's technology will break the sentence down into a Keynet structure it can better use to find search results: "emotional -> side effects -> [result_of] -> treatment -> [treats] -> colon cancer." This type of search engine will not just handle text. Jarg can use metadata to sift through images and multimedia files as well. Because of how the Jarg Knowledge Engine works, it can sift through literal millions of meta data entries to find the appropriate files. Is Jarg the wave of the future? It may well be, as context-oriented knowledge management is clearly going to be a great tool in sifting through all of the data amassed with each passing second. News Stories ICQ
Bot MyWeb
Creates Custom Portals with Bots Bot
Tool FIPA-OS v2 Released Bot
Contest Raises Stakes MetaRun
Provides True Web Partnerships Copernic
2001 Hits Shelves Kiwilogic
Triumphs at Internet World CoLIN Comet
Way Introduces Java Agent Kernel |
