‏إظهار الرسائل ذات التسميات ترجمة آلية. إظهار كافة الرسائل
‏إظهار الرسائل ذات التسميات ترجمة آلية. إظهار كافة الرسائل

السبت، 2 يونيو 2012

WIT3: a new collection of parallel texts!

We are very glad to announce WIT3 (to be read "wit cube"), acronym standing for Web Inventory of Transcribed and Translated Talks, a ready-to-use version of TED talks (wit3.fbk.eu).

TED is a non profit organization that invites experts to give the talk of their lives. TED records all talks and post them on its website (www.ted.com). Currently, more than 1100 talks are listed there, all subtitled in English. Translations of transcripts are also available into many languages (up to 90).

To make this collection of parallel texts more effectively usable by the MT research community, we have developed WIT3, a website hosting this multilingual corpus of talks, aligned at sentence level, alongside benchmarks, processing tools and reference MT results.

We hope WIT3 will offer an adequate service to the research community; for getting more info and downloading data, please visit: wit3.fbk.eu

الخميس، 24 مايو 2012

MEDAR MT package

The package includes:
  • Monolingual and parallel training data
  • Source corpora and their reference translations
  • Sentence and word alignment
    (morphosyntactic disambiguated)
  • An alignment package
  • Two MT baseline systems that uses MOSES
  • Evaluation metrics (BLEU/NIST and WER)
  • Formatting scripts to convert XML files
When downloading please sign the end-user agreement