Synchronous context-free grammars (SynCFG or SCFG; not to be confused with stochastic CFGs) are a type of formal grammar designed for use in transfer-based machine translation. Rules in these grammars apply to two languages at the same time, capturing grammatical structures that are each other's translations.
The theory of SynCFGs borrows from syntax-directed transduction and syntax-based machine translation, modeling the reordering of clauses that occurs when translating a sentence by correspondences between phrase-structure rules in the source and target languages. Performance of SCFG-based MT systems has been found comparable with, or even better than, state-of-the-art phrase-based machine translation systems. Several algorithms exist to perform translation using SynCFGs.
Formalism
Rules in a SynCFG are superficially similar to CFG rules, except that they specify the structure of two phrases at the same time; one in the source language (the language being translated) and one in the target language. Numeric indices indicate correspondences between non-terminals in both constituent trees. Chiang3 gives the Chinese/English example:
X → (yu X1 you X2, have X2 with X1)This rule indicates that an X phrase can be formed in Chinese with the structure "yu X1 you X2", where X1 and X2 are variables standing in for subphrases; and that the corresponding structure in English is "have X2 with X1" where X1 and X2 are independently translated to English.
Software
- cdec, MT decoding package that supports SynCFGs
- Joshua, a machine translation decoding system written in Java
References
Chiang, David (2007). "Hierarchical phrase-based translation". Computational Linguistics. 33 (2): 201–228. doi:10.1162/coli.2007.33.2.201. S2CID 3505719. https://doi.org/10.1162%2Fcoli.2007.33.2.201 ↩
Venugopal, Ashish; Zollmann, Andreas; Vogel, Stephan (2007). "An efficient two-pass approach to Synchronous-CFG driven statistical MT". Proc. NAACL HLT. pp. 500–507. https://www.aclweb.org/anthology/N07-1063 ↩
Chiang, David (2007). "Hierarchical phrase-based translation". Computational Linguistics. 33 (2): 201–228. doi:10.1162/coli.2007.33.2.201. S2CID 3505719. https://doi.org/10.1162%2Fcoli.2007.33.2.201 ↩