cmake -B build . -G "Unix Makefiles" && cmake --build build
# recommend -G Ninja instead

(prerequisites: GNU Make (3.8) C++17 capable compiler (GCC 8, or
visual studio 2017 will do) and [Boost](https://boost.org), which you
probably already have on your linux system; for Mac, you can get them
from [Homebrew](https://brew.sh/). Native windows builds should work;
you can also use cygwin or mingw.
### `make` options
If your system doesn't support static linking, `make NOSTATIC=1`
If you're trying to modify or troubleshoot the build, take a look at
`graehl/shared/graehl.mk` as well as `carmel/Makefile`; you shouldn't need to
manually run `make depend`.
## Subdirectories
* `carmel`: finite state transducer toolkit with EM and gibbs-sampled
  (pseudo-Bayesian) training
* `forest-em`: derivation forests EM and gibbs (dirichlet prior bayesian) training
* `graehl/shared`: utility C++/Make libraries used by carmel and forest-em
* `gextract`: some python bayesian syntax MT rule inference
* `sblm`: some simple pcfg (e.g. penn treebank parses, but preferably binarized)
* `clm`: some class-based LM feature? I forget.
* `cipher`: some word-class discovery and unsupervised decoding of simple
probabilistic substitution cipher (uses carmel, but look to the tutorial in
carmel/ first)
* `util`: misc shell/perl scripts