Machine Translation
Important Notions
- Translation as a noisy channel problem:
- Translation of source text S into
target text T:
argmaxT
P(T | S) = argmaxT P(T) P(S | T)
- Language model: P(T)
- Translation model: P(S | T)
- Alignment:
- Sentence alignment:
( S = [SS,1...SS,n] | T =
[ST,1,a(1)...ST,m,a(m)] )
- Word alignment:
( S = [wS,1...wS,n] | T =
[wT,1,a(1)...wT,m,a(m)] )
(where a(i) is the set of positions occupied by the
sentences/words in S that are aligned to the ith
sentence/word in T)
- Translation modeling:
- Translation: P(wS,i | wT,j,a(j))
(for i in a(j))
- Fertility: P(N | wT,j,a(j))
(where N is the cardinality of a(j))
- Distortion: P(i | j,m)
(for i in a(j))
- Language modeling
Slides for lecture 9
Suggested Reading
- Brown, P. F., Cocke, J., Della Pietra, S. A., Della Pietra, V. J.,
Jelinek, F., Lafferty, John D., Mercer, R. L. and Roossin, P. S.
(1993) The Mathematics of Statistical Machine Translation:
Parameter Estimation. Computational Linguistics 19, 263-312.
- Knight, K. (1997) Automating Knowledge Acquisition for Machine
Translation. AI Magazine 18(4).
- Manning, C. D. & Schütze, H. (1999) Foundations of Statistical
Natural Language Processing. MIT Press. Chapter 13.
Projects