Capers Jones has lots of stats on language and function points and
   productivity rates -- see WWWeb at SPR.Com and his text.

   As he would tell you -- its better to estimate the functionality from user
   descriptions via function points than backfiring your SLOCs from language
   A into FPs and then using his tables to estimate the boundaries of the
   projected conversion.  If you read his text  "Applied Software
   Measurement" he outlines various factors which can drive up line counts
   (post-COCOMO/Boehm).

   Also from my experience with COCOMO projections, the effort levels are
   also quite resposive (read hyper) to the environmental cost drivers -- in
   a new/adapted environment you cannot assume the same environmental
   factors.  Even from a small maintenenace task to larger conversion you
   must reconsider some of the expertise-level factors.

   -- Mike Berens
   Lockheed Martin