Search Descriptions

Main Topics

Search Publications


author

title

other

year

Pivot Languages

When parallel corpora for a specific language pair are rare or non-existent, then the bridging across a pivot language may be an option.

Pivot Languages is the main subject of 40 publications.

Publications

Machine translation systems for a language pair may still be built when resources in connection to a pivot (or bridge) language exist, for instance parallel corpora from each language into the pivot language (Callison-Burch and Osborne, 2003). This may take the simple form of building two machine translation systems, one into to pivot language, and the other from the pivot language to the target. Cettolo et al. (2011) show the effectiveness of this approach for Arabic-Italian via English. In a large-scale study, Koehn et al. (2009) show that pivoting through English often works as well as direct translation in the multilingual Acquis corpus. Pivoting works better if the bridge language is closely related to the other two languages (Babych et al., 2007). This allows Marujo et al. (2011) to leverage Brazilian Portuguese resources to improve an European Portuguese to English translation system.
It may be better to combine translation tables than simply tying machine translation systems together (Utiyama and Isahara, 2007). By using pivot languages in triangulation, existing phrase translation tables may be enriched, leading to better translation quality (Wu and Wang, 2007), especially when using multiple pivot languages (Cohn and Lapata, 2007). When constructing translation dictionaries via a bridge language (Tsunakawa et al., 2008), additional resources such as ontologies may help (Varga and Yokoyama, 2007).
Pivot languages have been used to improve word alignment, such as in the case of Chinese–Japanese, where only large parallel corpora of these languages paired with English exist (Wang et al., 2006). Bridge languages were also explored by Kumar et al. (2007).

Benchmarks

Discussion

Related Topics

New Publications

  • Miura et al. (2017)
  • Kholy and Habash (2014)
  • Kholy et al. (2013)
  • Hoang and Bojar (2015)
  • Dabre et al. (2015)
  • Miura et al. (2015)
  • Kholy et al. (2013)
  • Durrani and Koehn (2014)
  • Tsunakawa et al. (2013)
  • Zhu et al. (2013)
  • Zhu et al. (2014)
  • Dholakia and Sarkar (2014)
  • Dholakia and Sarkar (2014)
  • Kwon et al. (2013)
  • Nakov and Ng (2009)
  • Huck and Ney (2012)
  • Toral (2012)
  • Bertoldi et al. (2008)
  • Paul et al. (2009)
  • Habash and Hu (2009)
  • Wu and Wang (2009)
  • Do et al. (2010)
  • Leusch et al. (2010)
  • Zhang et al. (2010)
  • Khapra et al. (2010)
  • Schroeder et al. (2009)
  • Wen et al. (2010)
  • Saralegi et al. (2011)
  • Wu and Wang (2007)

Actions

Download

Contribute