Prototypes details

  • ALIGNER
  • A language independent aligner which breaks down parallel texts into smaller text segments. Each one of these segments should continue to be pairwise parallel. This aligner starts working, knowing nothing about the languages in which those parallel texts are written. At this stage of knowledge, our aligner enabled 75% precision for the language pair Portuguese-English (and 95% for the language pair Portuguese-Spanish) as shown in [Bilbao et alia, 2005]. Even, with this rate of misalignment it is possible then the extraction, either manually or automatically (using TREX), of translations of words and terms from misaligned parallel texts. Automatic extraction using no knowledge at all about the pair of languages, enables a precision close to 60%. This aligner was implemented by Tiago Ildefonso and is described at Ildefonso et al [2005]
  • 01 Dec 2005
  • Gabriel Pereira Lopes