This is a hybrid solution for acquiring the best translation out of multiple online MT engines

• THE BERKELEY PARSER - https://github.com/slavpetrov/berkeleyparser

  • You can find some grammar files here

• Query from KenLM - https://github.com/kpu/kenlm

• PHP with curl

• Java (for the Berkeley Parser)

• Berkeley Parser compatible grammar

• KenLM compatible language model (preferrably binarized)

• Access to at least two APIs

  • Google Translate - https://cloud.google.com/translate/
  • Bing Translator - http://www.bing.com/dev/en-us/translator
  • LetsMT - https://www.letsmt.eu

• Tokenized input sentences

• Google Translate
• Bing Translator
• LetsMT

The SyMHyT requires three parameters - the language model, input sentences, grammar file. It is run with the following command:

php SyMHyT.php <language model> <input sentences> <grammar>

For example:

php SyMHyT.php languageModel.binary inputSentences.txt eng_sm6.gr

The output generates four three files:

• output.google.txt
• output.bing.txt
• output.letsmt.txt
• output.hybrid.txt

The utils directory contains separate parts of the SyMHyT system that can be run as standalone

• utils/chunking/ contains files for individual chunking and unchunking

  • to parse an input file with the Berkeley Parser (a parsed file is required as input for the chunker) run:

   java -Xmx1024m -jar BerkeleyParser-1.7.jar -gr grammar.gr < input.txt
  

• utils/chunks_to_translated_chunks/ contains files for individual translating of chunked files

• utils/translated_chunks_to_hybrid/ contains files for running the hybrid system with chunked translated files