concerned with a theoretical analysis of LSI.
We derived an algebraic result about the singular value
decomposition, but it remained unclear what it meant for
the performance of LSI as an intelligent search technique.
In fact, this also remained unclear in the STOC'01 paper,
from which the result was taken.
In the meantime I see things clearer and will hopefully be
able to convey that to you.
In this context we will also talk about the various problems
and limitations of LSI (so far we had mainly praising).