Lecture 9

# Is l x f y hey gpk k1 goal compute the

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: put English sentence in y Let h() be the log probability under a tri-gram language model Let g() be a phrase pair score (from last slide) Then, the full translation score is: L X f (y) = h(e(y)) + g(pk ) k=1 §  Goal, compute the best translation y (x) = arg max f (y) y⇥Y(x) The Pharaoh Decoder §  Scores at each step include LM and TM The Pharaoh Decoder Space of possible translations §  Phrase table constrains possible translations §  Output sentence is built left to right §  but source phrases can match any part of sentence §  Each source word can only be translated once §  Each source word must be translated Scoring: §  In practice, much like for alignment models, also include a distortion penalty §  §  §  §  §  Define y = p1p2…pL to be a translation with phrase pairs pi Let s(pi) be the start position of the foreign phrase Let t(pi) be the end position of the foreign phrase Define η to be the distortion score (usually negative!) Then, we can define a score with distortion penalty: f (y) = h(e(y)) + L X k=1 g(pk ) + L 1 X k=1 ⇥ |t(pk ) + 1 §  Goal, compute the best translation y (x) = arg max f (y) y⇥Y(x) s(pk+1 )| Hypothesis Expansion 17 18 19 13 Hypothesis Expansion Hypothesis Expansion Maria Mary no no bofetada dio dio una bofetada una bofetada not give not give not did not no no did not give did not give a a slap slap a slap a slap slap slap a a la la bruja verde bruja verde bruja verde the the witch green witch green green witch green witch a la to to by by to the to the to to the the slap slap e: e: f: f: p: p: e: e: f: --------p: p: 1 • • witch witch -------*-------*.182 .182 e: f: p: Mary *-------.534 e: e: f: f: p: p: the witch the witch ... slap slap *-***---*-***---.043 .043 e: did not f: **------p: .154 e: slap f: *****---p: .015 e: the f: *******-p: .004283 e:green witch f: ********* p: .000271 Start with empty hypothesis ... e: no all foreign expansion Further hypothesis words Add another hypothesis covered – until English words ﬁnd foreign words that covers all foreign words – f: nobest hypothesiscover...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online