cdec is a fast decoder. SPEED COMPARISON ------------------------------------------------------------------------------ Here is a comparison with a couple of other decoders doing SCFG decoding: Decoder Lang. BLEU Run-Time Memory cdec c++ 31.47 0.37 sec/sent 1.0-1.1GB Joshua Java 31.55 2.34 sec/sent 4.0-4.8GB Hiero Python 31.22 27.2 sec/sent 1.7-1.9GB The maximum number of pops from candidate heap at each node is k=30, no other pruning, 3gm LM, Chinese-English translation task. GETTING STARTED ------------------------------------------------------------------------------ See the BUILDING file for instructions on how to build the software. To explore the decoder's features, the best way to get started is to look at cdec's command line options or to have a look at the test cases in the tests/system_tests/ directory. Each of these can be run with a command like ./cdec -c cdec.ini -i input.txt -w weights . The files should be self explanatory. EXTRACTING A SYNCHRONOUS GRAMMAR / PHRASE TABLE ------------------------------------------------------------------------------ cdec does not include code for generating grammars. To build these, you will need to write your own software or use an existing package like Joshua, Hiero, or Moses. OPTIMIZING / TRAINING MODELS ------------------------------------------------------------------------------ cdec does include code for optimizing models, according to a number of training criteria, including training models as CRFs (with latent derivation variables), MERT (over hypergraphs) to opimize BLEU, TER, etc. Eventually, I will provide documentation for this. ALIGNMENT / SYNCHRONOUS PARSING / CONSTRAINED DECODING ------------------------------------------------------------------------------ cdec can be used as an aligner. For examples, see the test cases. COPYRIGHT AND LICENSE ------------------------------------------------------------------------------ Copyright (c) 2009 by Chris Dyer Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. The LBFGS implementation contains code from the Computational Crystallography Toolbox which is copyright (c) 2006 by The Regents of the University of California, through Lawrence Berkeley National Laboratory. For more information on their license, refer to http://cctbx.sourceforge.net/