1CS 294-5: StatisticalNatural Language ProcessingPhrase-Based TranslationLecture 12: 10/17/05Uses slides from Koehn, Knight et al.Phrases in IBM Modelsil hoche la têtehe is noddingPhrase-Based Systems Pharaoh’s Model[Koehn et al, 2003]Segmentation Translation DistortionPharaoh’s ModelWhere do we get these counts?Extracting Phrases2Phrase Size Phrases do help But they don’t need to be long Why should this be?Bidirectional AlignmentAlignment Heuristics Sources of AlignmentsLexical Weighting The Pharaoh Decoder Probabilities at each step include LM and TM3Hypotheis Lattices Pruning Problem: easy partial analyses are cheaper Solution 1: use beams per foreign subset Solution 2: estimate forward costs (A*-like)What’s Next? Modeling syntax PCFGs and phrase structure Syntactic parsing Grammar induction Syntactic language and translation models Speech systems Acoustics
View Full Document