Perl has PDL which has a rich library through CPAN, and of course there's matlab, octave, mathematica, R, and so on.
EDIT: Just double checked, and Perl's PDL does indeed integrate with LAPACK via PDL::LAPACK, and with BLAS via PDL::LinearAlgebra, PDL::LinearAlgebra::Real, and PDL::LinearAlgebra::Complex.
I suppose the real question here is, if you're looking to doodle around with some ideas you can do it in just about any language. The problem with using other languages is that you begin to fight the language itself when dealing with less than ideal data (e.g. the ever-present ill-conditioned matrix).
I had a Numerical Analysis professor way back in the day that hated anything but FORTRAN because languages like C would flip between single and double precision calculations without warning. (I know it's much more well-defined now, but back then gcc was just a blip and ANSI was just a dream).
My suggestion is to use NumPy or Perl's PDL if you need to also tie into system (e.g. databases, web interfaces, etc), or use R/Matlab/Mathematica if you're closer to "pure" math, otherwise fall back to the standards Fortran and/or C.