WING presentation

1
WING presentation

• Hendra Setiawan
• Wed, 3rd August 2005

2
A brief illustration about SMT
Champollion
SMT System
Egyptian hieroglyph
Rosetta Stone
unknown sentence
parallel texts
3
Challenges in SMT

• Lexical Ambiguity
• Book the flight reservar
• Read the book libro
• Differing word orders
• English word order is subject-verb-object
• Japanese word order is subject-object-verb
• Syntactic structure is not preserved across
  translations
• The bottle floated into the cave la botella
  entro a la cuerva flotando (the bottle entered
  the cave floating)
• Pronoun resolution
• Differing Treatments of tense across languages

4
Noisy Channel Model

• IBM Model Brown 1993

5
Joint Source Channel Model

• Originally proposed by Li and Zhang 2004 for
  machine transliteration task

• Phrase Translation Learning through LOD
  Processing (Hendra, et. al 2005)

• Augment JSCM model to incorporate target
  permutation through n-gram
• statistics (join with Zhang 2005)

• Improve the framework through better modeling of
  lexical mapping and
• word re-ordering

6
Comparison

• Joint Source Channel Model

• Noisy Channel Model

• Missing!

7
Noisy Channel Model (IBM)
8
Joint Source Channel Model

• Translate Tommorow I will fly to to the
  conference in Canada to German

• Re-ordered