Search Descriptions

General

Neural machine Translation

Statistical Machine Translation

Search Publications


author

title

other

year

Other Approaches to Machine Translation

While the statistical approach currently dominates research in machine translation, many other approaches have a rich history. Reaching back to these approaches motivates combining divergent architectures into hybrid systems.

Other Approaches is the main subject of 57 publications. 21 are discussed here.

Topics in Introduction

Research Groups | Available Software | Speech Translation | Sign Language | Computer Aided Translation | Other Approaches | Deployment

Publications

Statistical machine translation is related to other data-driven methods in machine translation, such as the earlier work on example-based machine translation (Somers, 1999; Hutchins, 2005). Contrast this to systems that are based on hand-crafted rules (Nirenburg et al., 1992; III and Mitamura, 1992).
The distinctions between these categories are getting blurred. Borrowing from different approaches is called the hybrid approach to machine translation. Statistical methods, especially the use of the language model, may be integrated in rule-based systems (Knight et al., 1994; Knight et al., 1995; Habash and Dorr, 2002). Parsing and translation decisions may be learned from data (Hermjakob and Mooney, 1997). Multiple scoring functions may decide between the alternatives generated by the transfer-based system (Carl, 2007).
Statistical methods may be used to learn rules for traditional rule-based systems (Llitjós and Vogel, 2007; Wolf et al., 2011). Conversely, translations from rule-based systems may be used as additional phrase translations in statistical systems (Chen et al., 2007). Rule-based systems may be used to generate training data for statistical methods (Hu et al., 2007), essentially having the statistical method relearn the rule-based system (Dugast et al., 2008).
Statistical machine translation may serve as a fall-back for methods that frequently fail to produce output, but are more accurate when they do Chai et al. (2006).
Statistical machine translation models may be also used to automatically post-edit the output of interlingual (Seneff et al., 2006) or rule-based systems (Simard et al., 2007; Simard et al., 2007b). Additional markup from the rule-based system may be exploited for tighter integration (Ueffing et al., 2008). Such post-editing may alleviate the need to customize rule-based systems to a specific domain (Isabelle et al., 2007). Labaka et al. (2007) compare different machine translation approaches for the Basque–Spanish language pair.

Benchmarks

Discussion

Related Topics

The hybridization of systems that follow different machine translation approaches may also take the form of system combination. Ideas from other approaches to machine translation are also incorporated into syntax based models.

New Publications

  • Costa-jussà et al. (2013)
  • Ney (2013)
  • Hoang et al. (2014)
  • Richardson et al. (2014)
  • Lanctôt (2014)
  • Bond et al. (2011)
  • Way (2010)
  • Zbib et al. (2012)
  • España-Bonet et al. (2011)
  • Sánchez-Cartagena et al. (2011)
  • Ahsan et al. (2010)
  • Galron et al. (2009)
  • Babych et al. (2012)
  • Dandapat et al. (2012)
  • Federmann (2012)
  • Harriehausen-Mühlbauer and Heuss (2012)
  • Kirkedal (2012)
  • Tambouratzis et al. (2012)
  • Bojar et al. (2005)
  • Federmann et al. (2012)
  • Hunsicker et al. (2012)
  • Molchanov (2012)
  • Caseli et al. (2006)
  • Hanneman et al. (2009)
  • Wehrli et al. (2009)
  • Xu and Seneff (2008)
  • Lagarda et al. (2009)
  • Dugast et al. (2009)
  • Thurmair (2009)
  • Varga and Yokoyama (2009)
  • Federmann et al. (2010)
  • Habash et al. (2009)
  • Aleksic and Thurmair (2011)
  • Groves and Way (2005)
  • Surcin et al. (2007)
  • Nakazawa et al. (2006)

Actions

Download

Contribute