Translating%20Collocations%20for%20Bilingual%20Lexicons

1
Translating Collocations for Bilingual Lexicons

• Collocations (idiomatic multi-word expressions)
  difficult to translate
• semantically opaque
• cannot be translated word-by-word
• a major obstacle to second language acquisition
• Example demonstrate support ? prouver son
  adhésion (prove adherence)

2
The Champollion approach

• Input Large parallel corpora
• Output List of collocations in each language,
  and equivalence mappings between these
  collocations
• The method is statistical and language-independent

3
Algorithm

• Align sentences across corpora
• Extract collocations from co-occurrence
• Identify all words that frequently appear across
  a source collocation
• Iteratively consider and score combinations of
  those words
• Select best set of words for the translation
• Determine word order and fill in prepositions

4
Sample translations

• additional costs ? coûts supplémentaires
• affirmative action ? action positive
• free trade ? libre-échange
• freer trade ? libéralisation échanges
• take steps ? prendre mesures
• stock market ? bourse

5
Evaluation results

• Corpus of 3.5 million words, collocations
  selected from the same corpus 78
• Corpus of 8.5 million words, collocations
  selected from the same corpus 74
• Corpus of 3.5 million words, collocations
  selected from a different corpus 65

6
Conclusion

• Champollion provides for collocation translation
• Robust
• Language-independent
• Requires no tools
• But Requires parallel corpora