Search Descriptions


Neural machine Translation

Statistical Machine Translation

Search Publications





Word Alignment Combination

System combination has been a very successful method for statistical machine translation, so the idea has been also picked up to improve word alignments.

Word Alignment Combination is the main subject of 16 publications. 7 are discussed here.


Tufiş et al. (2006) combine different word aligners with heuristics. Schrader (2006) combines statistical methods with manual rules. Elming and Habash (2007) combine word aligners that operate under different morphological analysis of one the languages, in their case Arabic. Huang et al. (2005) discuss the issue of interpolating word alignment models trained on data from multiple domains. Word alignment in a specific domain may also be improved with a dictionary obtained from a general domain (Wu and Wang, 2004). Combining different word aligners that stem from different methodologies may also improve performance (Ayan et al., 2004). The log-linear modeling approach may be used for the combination of word alignment models with simpler models, for instance based on part-of-speech tags (Liu et al., 2005).



Related Topics

System combination of output from different machine translation systems is a more complex problem, but has been shown to be very successful.

New Publications

  • Liu et al. (2013)
  • Tu et al. (2011)
  • Tu et al. (2012)
  • Pal et al. (2013)
  • Xu and Rosti (2010)
  • Deng and Zhou (2009)
  • DeNero and Macherey (2011)
  • Xi et al. (2011)
  • Tufiş et al. (2005)