MBOT Moses Translation System
Fabienne Braune, Nina Seemann, Daniel Quernheim, Andreas Maletti
The MBOT Moses Translation System is a statistical machine translation system based on multi bottom-up tree transducers (MBOT). An MBOT is a synchronous grammar that models discontinuities by allowing a sequence of target tree fragments on the right hand side of its rules. A formal description can be found in .
We implemented various rule extractions to test the power of MBOT in different settings:
- minimal tree-to-tree rules as described in  and 
- non-minimal string-to-tree rules as described in 
- non-minimal tree-to-tree and string-to-string rules as described in 
- non-minimal tree-to-string rules
Furthermore, we developed a method that allows one to transform dependency parse trees into constituent-like tree representations. The method and evaluation is given in . The dependency parser we used yielded sometimes non-projective dependency parse trees. Those need to be converted into projective ones. If you are interested in this code, please contact Anders Björkelund [Link].
- Andreas Maletti. How to Train your Multi Bottom-up Tree Transducer. Proc. 49th ACL, 2011.
- Fabienne Braune, Nina Seemann, Daniel Quernheim & Andreas Maletti. Shallow Local Multi Bottom-up Tree Transducers in Statistical Machine Translation. Proc. 51st ACL., 2013.
- Nina Seemann, Fabienne Braune & Andreas Maletti. String-To-Tree Multi Bottom-up Tree Transducers. Proc. 53rd ACL, 2015.
- Nina Seemann, Fabienne Braune & Andreas Maletti. A Systematic Evaluation of MBOT in Statistical Machine Translation. Proc. MT Summit XV, 2015.
- Nina Seemann & Andreas Maletti. Discontinuous Statistical Machine Translation with Target-Side Dependency Syntax. Proc. 10th WMT, 2015.
- Download the latest version of the decoder from git:
git clone -b mbotTestedDecoder git://github.com/moses-smt/mosesdecoder.git
- minimal tree-to-tree: Code for rule extraction, lexical scoring, etc. Instructions on training and tuning. [download]
- non-minimal tree-to-tree, string-to-tree, tree-to-string, and string-to-string: Instructions on preprocessing, training and tuning. [download]
- Instructions and code for transforming dependency parse trees. [download]