Faster querying with smaller memory usage.

DALM is faster and smaller language model implementation using double-array structures.

Quick Start

# docker pull nowlab/dalm_translation_kit:20160330

You may see DALM components and Moses decoder under /opt. Moses decoder is linked to DALM.

Build DALM from source.

C++

# mkdir build && cd build && cmake .. -DCMAKE_INSTALL_PREFIX=../local && make install

(Experimental) Python

(comming soon...)

Use DALM with your language model.

You can use DALM with your language model. DALM can read ARPA format.

Build binary model

# build_dalm -f /path/to/arpa.file -o /path/to/output

Use DALM with Moses decoder.

DALM is integrated with Moses decoder. We also distribute a docker image including pre-build DALM and Moses decoder. >> Our Dockerhub page

Build Moses decoder with DALM

# ./bjam --with-dalm=/path/to/DALM

Write moses.ini

[feature]
...
PhraseDictionaryMemory name=TranslationModel0 num-features=4 path=/path/to/rule-table.gz input-factor=0 output-factor=0
...
DALM name=LM0 factor=0 path=/path/to/dalm_model_directory order=5

...
[weight]
LM0= ...

Use DALM with your code.

From C++

Please see an example.

(Experimental) From Python

Please see an example.

License