Search Descriptions


Neural machine Translation

Statistical Machine Translation

Search Publications





Factored Translation Models

Factored translation model use a richer representation of words, a vector of factors, instead of the typical encoding of words as unique tokens.

Factored Translation Models is the main subject of 17 publications. 8 are discussed here.


Factored translation models (Koehn and Hoang, 2007) allow the integration of syntactic features into the translation and reordering models. Already earlier work augmented phrase translation tables with part-of-speech information (Lioma and Ounis, 2005). The approach has been shown to be successful for integrating part-of-speech tags, word class factors (Shen et al., 2006), CCG super-tags (Birch et al., 2007), and morphological tags (Badr et al., 2008) for language modeling. Complementary, the source side may be enriched with additional markup, for instance to better predict the right inflections in a morphologically richer output language (Avramidis and Koehn, 2008). More complex factored models for translating morphology have been explored for English–Czech translation (Bojar, 2007).
Additional factors may also be considered in the language model. Niehues et al. (2011) include aligned source words in the target word representation which enables so-called bilingual language models.



Related Topics

New Publications

  • Huet et al. (2013)
  • Singh and Bandyopadhyay (2010)
  • Singh and Bandyopadhyay (2010)
  • Wang et al. (2012)
  • Bojar et al. (2012)
  • Koehn and Haddow (2012)
  • Cettolo et al. (2008)
  • Yeniterzi and Oflazer (2010)
  • Graham and van Genabith (2010)
  • Rishøj and Søgaard (2011)