Guiding Statistical Word Alignment Models With Prior Knowledge

Author : Yonggang Deng and Yuqing Gao

Keypoint:

• Human knowledge can significantly improve word alignment F-measure and translation performance
• The morereasonable constraints are imposed ,the easier the task would become
• Proper names with low counts then are encouraged to link to proper names during training
• Putting reasonable constraints learned from monolingual analysis can alleviate data spareness problem in bilingual applications
• Measure translation performance by theBLEU score (Papineni et al., 2002) and TranslationError Rate (TER) (Snover et al., 2006)
• Constraints improve word alignmentprecision and accuracy of phrase translation tablesas well

Relative Link:

• LSA :
  en.wikipedia.org/wiki/Latent_semantic_indexing
• SVD:
  en.wikipedia.org/wiki/Singular_value_decomposition
• MLE:
  en.wikipedia.org/wiki/Maximum_likelihood