The e-SMT module performs Statistical Machine Translation (SMT) using DKT-Project-specific wrapper scripts built on top of Moses, an open-source SMT system written in C++ (http://www.statmt.org/moses/). Given a trained model, new input can be translated and displayed in NIF format.
Note, the current implementation translates in the following four directions:
- German to English
- English to German
- Spanish to English
- English to Spanish
The system is constantly being updated and newer models as well as language pairs will be added (Arabic-English).
https://api.digitale-kuratierung.de/api/e-smt
The API conforms to the general NIF API specifications. For more details, see: http://persistence.uni-leipzig.org/nlp2rdf/specification/api.html In addition to the input and Content-type (header) parameters, the following parameters have to be set to perform Statistical Machine Translation on the input:
source-lang: The language of the input text.
target-lang: The language of the output text.
For now, only the following language pairs are supported. English (en), German (de), and Spanish (es)
A document in NIF format.
Example cURL post for e-smt:
curl -X POST --header "Content-type:text/plain" -d 'Click on the right button.' "https://api.digitale-kuratierung.de/api/e-smt?source-lang=de&target-lang=en"
This end-point (hosted on the dkt-api server) relies on external software and assumes that Moses (the MT engine) is installed at: /usr/local/mt/bin, and the handler script (translate_main.sh) is placed at: /usr/local/mt
Moses and its dependencies can be locally installed, following the instructions at http://www.statmt.org/moses/?n=Development.GetStarted. Please contact Ankit Srivastava (firstName.lastName@dfki.de) for more information.