[Spambayes] Re: egregious patents on anti-spam techniques

[Gary Robinson]

> 
> Graham derived filters do map words into a 'continuous semantic vector
> space', namely the one dimensional vector
> space of the range of [0.0, 1.0] of real numbers, and then 'clustering
> techniques' are applied. Normally clusters are
> defined by hyperplanes in N-Space, but in one dimesion they would be
> threshold values. The two 'symantic anchors' are arguably cluster
> centers located at 0.0 and 1.0 (also known as ham and spam in
> Graham-derived filters).
> 


I completely understand what you're saying, but having spend a lot of money
and time researching the doctrine of equivalents and the various court
rulings about it, I really, really think that's too much of a stretch. To be
equivalent, the method has to perform "substantially the same function in
substantially the same way"  as the one in the patent. (Graver Tank & Mfg.
Co. v. Linde Air Products, 339 U.S. 605 (1950))

The latter requirement is just not the case here, IMO.

Moreover even if it were the case, many things can happen during the
prosecution of a patent that make the doctrine of equivalents unavailable to
the patent holder. In order to get a grasp on that in this specific case, we
would have to look at the "file wrapper" which contains the history of the
interaction with the patent office examiner. The wind in the courts has been
blowing against the doctrine of equivalents for years.

And without the DoE, you have to infringe exactly, and that is not the case
here. 


> In fact it is quite reasonable to describe a Graham-derived filter as
> having a 'ham anchor' that can
> be described as a location in N-Space in which each token string
> describes a dimension and the
> 'clue' value for that string is the location of the anchor in that
> dimension.

Again, that is just way too much of a stretch to invoke the DoE IMO. There
is no point in n-space that represents spam or ham in Graham. There is no
location. We are reducing everything to one dimenension before we ever try
to determine ham or spamminess, and being near 0 or 1 is not "substantially
the same" as measuring a distance to a point in n-space.

One thing I don't think we should do though is get into a nit-picking
argument about this. It isn't worth it. :)

The main thing is that I agree, prior art should be found if possible to
lessen the chance that anybody is hassled with this. And of course there is
always a random chance element when things actually get to trial. And
getting things right can involve an expensive appeal process. I've read
about very stupid decisions in the lower courts that were corrected on
appeal. But the open-source community wouldn't easily be able to pay for
such a long and expensive process.

So it's certainly better to be safe than sorry.

> I'd like to suggest that it would be good to file a protest as Kaitlin
> suggested. There was certainly
> work done in this area before June 14, 2001.  Does anyone have pointers
> they can pass along.

This might help and couldn't hurt. As I read the claims and specification,
the claims are specific to a particular methodology that is different from
Graham and so I doubt that prior art that is graham-like will overturn the
claims. However, the prior art may go into the prosecution history in such a
way that it could greatly reduce any chance that anyone could try to use the
patent to attack graham-based filters. So it would be worth doing if anyone
can find some prior art.

Gary