If you saw my earlier post about Wolfram Alpha, then you know that semantic search has been on my mind recently. To define it very broadly, a semantic search engine is one that understands human language as it is actually used. Current generation search engines operate by looking for keywords - we've all essentially trained ourselves to ask questions in a way that search engines understand. The challenge with this is that sometimes we're looking for something a bit too fuzzy for keyword-based search.

A good example of this is image searching. Suppose you were looking for a screencap from the scene in The Godfather where Don Corleone collapses in his tomato garden. Google Image Search is tremendously powerful, but it's keyword-based, so searching for this on GIS would entail a search string like "godfather collapse" and then trawling through the results until you find what you're looking for. An ideal semantic search engine could just be asked "Where can I find a screencap from the scene in The Godfather where Don Corleone collapses in his tomato garden?", and it would know precisely what you want. There's a lot of words in that query that are useless to a keyword search engine but that are essential for a human (or a semantic engine) to understand what you're after.