Search Patents

Software patents are the kiss of death for innovation, because under current practice you can get the most obvious 'method' for doing just about anything patented. Have you ever swung sideways on a swing? Patent infringer!

Who can afford to litigate dubious patents? Certainly not individual researchers or developers. Bruce Perens has a great essay on the chilling effects of bogus patents in the software world.

The one and only time software patents give you a fuzzy feeling is when you stumble across prior art.

For instance, anyone who is interested in trying latent semantic analysis on text collections, but has been scared off by patent #4,839,853 (June, 1989) might be interested in the following:

A possible approach to optimal indexing for a static collection could result from the application of principal component analysis. This would result in the determination of a set of axes selected to optimize the variance of the document projections on each axis. Each document wold then be 'indexed' by its projections on the various axes, and the query would be similarly transformed. This technique has not been studied in the context of information retrieval..."

That's a quote from Scott Preece's PhD Thesis (say it again, make it rhyme!), published eight years before the patent claim was filed. (October, 1981). And latent semantic analysis is nothing but a certain kind of component analysis where the components are constrained to be orthogonal.

What does it mean?

It means old Mr. Preece had a big brain. And it means that you can build a search engine employing nineteenth-century linear algebra without having to worry about a lawsuit.

Unless, of course, you do something crazy like parse your search query. Because of course that's patented.