Search Descriptions

Main Topics

Search Publications





Tree to String Models

Tree to string models use rich source side language representation to translate into unannotated word sequences in the target language.

Tree To String is the main subject of 33 publications.


Similarity between the source syntax tree and the derivation tree during decoding may be used as a soft constraint (Zhou et al., 2008). Similarly, nonterminals in rule applications that match the source syntax tree could be flagged in a feature (Marton and Resnik, 2008).
Syntax-directed translation parses the input first and uses the parse structure to guide the output phrase generation (Huang et al., 2006; Huang et al., 2006b). Liu and Gildea (2008) extend this work by adding semantic role labels and changing parameter estimation.
A simpler method follows the phrase-based approach but informs phrase selection with shallow source trees (Langlais and Gotti, 2006). Given the segmentation into phrases and a source syntax tree, both the structural changes to the tree to fit the phrases and reordering of the restructured phrase tree may inform better translations (Nguyen et al., 2008).
Building on the hierarchical phrase based model, Zhang et al. (2007) use syntactic annotation of the input to make a distinction between linguistic and non-linguistic phrases. Hopkins and Kuhn (2007) present a tree-to-string model with best-first search.



Related Topics

New Publications

  • Neubig (2013)
  • Li et al. (2015)
  • Hanneman and Lavie (2013)
  • Xie et al. (2014)
  • Li et al. (2014)
  • Cmejrek et al. (2013)
  • Huang et al. (2013)
  • Meng et al. (2013)
  • Cmejrek (2014)
  • Neubig and Duh (2014)
  • Neubig (2013)
  • Mi et al. (2010)
  • Zhai et al. (2012)
  • Hassan et al. (2009)
  • Feng et al. (2012)
  • Zhang and Chiang (2012)
  • Lee (2012)
  • Liu and Gildea (2009)
  • Zhang et al. (2009)
  • Wu et al. (2010)
  • Hoang and Koehn (2010)
  • Liu et al. (2011)
  • Zhang et al. (2011)
  • Xie et al. (2011)