Semi-Automatic%20Learning%20of%20Transfer%20Rules%20for%20Machine%20Translation%20of%20Low-Density%20Languages

1
Semi-Automatic Learning of Transfer Rules for
Machine Translation of Low-Density Languages

• Katharina Probst
• Language Technologies Institute
• Carnegie Mellon University

2
Introduction and Motivation
3
Low-Density Languages
4
Elicitation Corpus

• Designed for bilingual informant, who provides
  translations and word alignments
• Aimed at covering major typological linguistic
  features (Swadesh list)
• Organized in a compositional way e.g. NPs first,
  then PPs, Ss, etc.
• Goal feature detection and diagnostic planning
  to limit the number of sentences the user
  translates

5
Seed Generation

• Learn rules from sentence pairs
• Abstracted to the POS level, but still specific
  to the sentence
• Seed rules are first guesses at useful transfer
  rules

6
Compositionality
7
Seeded Version Space Learning
8
Conclusions
9
Future Work
10
(No Transcript)