Flattened Syntactical Phrase-Based Translation Model for SMT

  • Authors:
  • Qing Chen;Tianshun Yao

  • Affiliations:
  • Natural Language Processing Lab, Northeastern University, Shenyang, P.R. China;Natural Language Processing Lab, Northeastern University, Shenyang, P.R. China

  • Venue:
  • ICCPOL '09 Proceedings of the 22nd International Conference on Computer Processing of Oriental Languages. Language Technology for the Knowledge-based Economy
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposed a flattened syntactical phrase-based translation model for Statistical Machine Translation (SMT) learned from bilingual parallel parsed texts. The flattened syntactical phrases are sets of ordered leaf nodes with their father nodes of single syntax trees or forests ignoring the inner structure, containing lexicalized terminals and non-terminals as variable nodes. Constraints over the variable nodes in target side guarantee correct syntactical structures of translations in accordant to the syntactical knowledge learned from parallel texts. The experiments based on Chinese-to-English translation show us a predictable result that our model achieves 1.87% and 4.76% relative improvements, over Pharaoh, the state-of-art phrase-based translation system, and the system of traditional tree-to-tree model based on STSG.