About Śata-Anuva̅dak Translation Resources

Anoop Kunchukuttan, Abhijit Mishra, Rajen Chatterjee, Ritesh Shah, Pushpak Bhattacharyya. 2014. Śata-Anuva̅dak: Tackling Multiway Translation of Indian Languages . Language and Resources and Evaluation Conference (LREC 2014). 2014.

What can you do with these resources?

• Translate sentences using the Moses decoder
• Improve the translation systems by developing post-editing methods
• Build Pivot based SMT systems by harnessing multiple SMT systems. (See Pivot language approach for phrase-based statistical machine translation )
• Build translation applications like a mobile app, multi-lingual chat, etc.

Instructions for Use

• For each language, a 5-gram language model in APRA format can be downloaded.
• For each language pair, the phrase table, reordering model and moses.ini file for the tuned model can be downloaded.
• The systems have been trained using Moses 1.0. Since Moses 2.0, the format of moses.ini has changed, so you may want to use Moses 1.0 for translation. Alternatively, Moses provides a script to convert v1 format files to v2 (see: mosesdecoder/scripts/training/convert-moses-ini-to-v2.perl)
• In the moses.ini file, edit the paths to the language model, phrase table and the reordering model before using these models in Moses:
  • Language Model: Under Section [lmodel-file], field number 4
  • Phrase Table: Under Section [ttable-file], field number 5
  • Reordering Model: Under Section [distortion-file], field number 4
• Known Issues:
  • Some phrase tables contain the characters '[' and ']'. The latest Moses decoder fails to decode such phrase tables. The issue can be resolved either by escaping these characters or deleting them from the phrase tables.

Contact

License

Download

Language Models:

Source Reordered, Phrase Based Translation Models

Phrase Based Translation Models