4/1/99 
Attendees: Rajeev, Svetlozar, Sebastien, Shinji, Yue

We spent most of the meeting time talking about more SMILE implementation
issues.  The new implementation idea is to 'compress', or 'token-lize' the
input document before looking into nested structures. The compression may
give two advantages: the efficient reading of input files and the
avoidance of ambiguity when detecting nested structures. Svetlozar also
pointed out that we need to add more variety and meaning tags to the
converted documents to facilitate the classification algorithm.