NLP - Machine Translation

Challenges

• Lexical Ambiguity
• Differing Word Orders
• Syntactic Structure is not Preserved
• Syntactic Ambiguity
• pronoun resolution

Classical MT (rule-based)

• Direct MT
  – word by word, no syntactic or semantic analysis
  – long-range reorderings
• transfer-based approaches
  – analysis, transfer, generation
• interlingua based translation
  – analysis into lang independent representation & generation
  – what is representation: intersection of breaking concepts

Statistical MT

• parallel corpus
• Noisy Channel Model
  – $argmax_ep(e)p(f | e)$
• IBM model 1
  – alignment a: $(l+1)^m$
  – $a* = arg max_a p(a | f,e,m)$
  – p(f, a | e,m)= p(a | e,m) p(f | a,e,m)
  – p(f | e,m)
  – translation prob: $p(f | a,e,m)=\prod_{j=1}^mt(f_j|e_{a_j})$
  – the generative process: $p(f ,a|e,m)=p(a | e,m) p(f | a,e,m)=\frac{1}{(l+1)^m}\prod_{j=1}^mt(f_j|e_{a_j})$