Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
README-295 | 2010-05-12 | 1.1 kB | |
cmu-ebmt-2.95.zip | 2010-05-12 | 2.0 MB | |
Totals: 2 Items | 2.0 MB | 4 |
New version 2.95. The main programs are pretty vanilla C++, but there are also a bunch of C-shell scripts. The system is known to work on Linux and should work on any *nix system (BSD, Solaris, etc.) with minimal tweaking. The main programs used to run under Windows, compiled with OpenWatcom (www.openwatcom.org), but that hasn't been tested for a while; you'll also need Cygwin if you want to run any training or tuning under Windows (all data files are architecture independent, so you can train a system under Linux and then just copy everything but the executables over to Windows). CHANGES SINCE 11/2009 RELEASE: Added parameter tuner. Revamped scoring to use an explicit feature vector rather than a series of member variables, which makes it much easier to add new features. Added a bunch of new features to the log-linear model. Removed a bunch of obsolete code and settings. This release is all in one archive, and includes a sample translation system for English to Haitian Creole, with instructions for building the corresponding Haitian Creole to English system.