English-French aligned bitexts extracted from Wikipedia

These data were extracted from the English and French Wikipedia dump dated May 23, 2011. The extraction method is presented in the master thesis of Lise Rebout entitled "L'extraction de phrases en relation de traduction dans Wikipédia" directed by Philippe Langlais.

  • Aligned data used for training, development and testing of classifiers and machine translation systems:
    train+dev+test.txt (448 kB)
    2 058 lines, a pair of sentences per line. The French phrase is separated from the English sentence by a tab.
  • Pair of French-English sentences extracted using classifiers and a bootstrapping phase:
    bitexteFR-EN.txt.bz2 (32 MB)
    665 297 lines, a pair of sentences per line. The French phrase is separated from the English sentence by a tab.

    These sentence pairs constitute a quasi-parallel corpus. The bitext are not exactly the translation of each other. The parallelism rate is about 81%.

For questions, please contact Lise Rebout.