markovian matching & dobly noise reduction

[Boris 'pi' Piwinger]

Tom Anderson <tanderso at oac-design.com> wrote:

>> Also I find it absurd to say a filter is better than a
>> human. Who checked the filter after all?
>
>Slashdotters had that discussion too.  When they say better than a
>human, they mean that a human scanning 5000 emails is bound to skip some
>legitimate ones due to sheer volume and the therefore scan rate.  

That might be true, but also true for checking the filter. I
don't trust people who try to fool you with this kind of
statistics ...

>You
>simply can't look closely at all of them.  Only when you look closely
>can you achieve 100% accuracy.  Moreover, they specifically say that by
>a human they mean a personal secretary rather than yourself, and

... and then blame someone else.

>therefore they will not know your exact desires even looking closely. 
>Using these criteria, I find it very likely that a human will do even
>more poorly than 99.5%.  

That would be one error in 200. 

>Think about the number of typos you see in
>published works (or bugs in software) to judge human fallibility. 

That is a completely different level. That is if you compare
if a human can tell a car from a bike (telling ham from
spam). And you say: Humans are not good at it because they
cannot say it that particular brand was build in 1993 or
1994.

>Automated spell-checkers _can_ be more accurate too (rhetorically
>speaking, please).

And they can really destroy texts. Examples are discussed
with great fun;-)

pi