carmel: finite state transducer toolkit (and k-best graph paths).

 http://www.isi.edu/licensed-sw/carmel/

 EM and gibbs-sampled (pseudo-Bayesian) training
 (see carmel/README and carmel/carmel-tutorial).
 To build, cd carmel; make install
 (see carmel/LICENSE - free for research/non-commercial)

forest-em: derivation forests EM and gibbs (dirichlet prior bayesian) training

shared: generic routines used by both

PREREQUISITES:

Boost (http://boost.org - usually already installed in /usr)

BUILDING:

If this directory is not called 'graehl', then either rename it, or make sure
there's a symlink from . to graehl (ln -s . graehl) in this directory.  This is
necessary for the headers <graehl/shared/X> to be located - or you can install
the headers with

    ./install.sh /usr/local

To specify where to install built binaries, do for e.g. Carmel:

    cd carmel
    INSTALL_PREFIX=/usr/local ARCH_FLAGS="-m32" make -j 4 install

or, to build without installing (look in bin/`hostname`for binaries)

    make -j 4

You can have the binaries built in bin/YOURDIR by:

    make -j 4 BUILDSUB=YOURDIR

If your system can't build static executables, you can:

   make -j 4 NOSTATIC=1

Static executables are already excluded if ARCH=macosx

Requires GNU make 3.80 or later, and a recent C++ compiler (gcc 4.3 is
known good, as is Microsoft Visual C++ 7.1)

If you're mystified by what's going on in the build, look at shared/graehl.mk

"make depend" should run automatically.