Named Entity Recognition (NER) How does Snowball vector weighting work?

I'm interested in extracting named entities with Snowball.

Eugene Agichtein has a great presentation on how Snowball does it at http://www.mathcs.emory.edu/~eugene/talks/dl00.ppt

Can someone explain how the vector weighting of the 5-tuples work in Snowball.

A Snowball pattern vector is a 5-tuple <left, tag1,="" middle,="" tag2,="" right="">, tag1, tag2 are named-entity tags left, middle, and right are vectors of weighed terms.

For example the middle vector could be a vector with these weights: {<'s 0.5>, <central 0.5=""> <headquarters 0.5="">, < in 0.5>}

Is each word weighted (via Log or TFIDF) according to the current sentence?