This paper is concerned with the evaluation of Hyperlog.  We first define
the semantics of Hyperlog programs in terms of a bottom-up `naive'
evaluation strategy.  We then show the semi-naive evaluation algorithm
developed for relational languages can be adapted for the more efficient
evaluation of Hyperlog programs.  Since we only have one <I>graph</I>
predicate, this is achieved by syntactically partitioning the templates of
rules into those templates which cannot be affected by any updates to the
database during the evaluation of the program and those templates which may
be affected by updates to the database.  The former templates are the
analogue of EDB predicates and the latter of IDB predicates in relational
rules.