Skip to content

Latest commit

 

History

History
 
 

week06_mt

Machine Translation: Word alignment models, noisy channel models, phrase-based MT, NMT etc.

Today's lecture will cover some of the key models that have been applied to Machine Translation in the past 20 years and explain some of the challenges posed by this task.

Building on the EM algorithm from last week, we will look at how word alignments can be inferred from parallel text, how early statistical approaches built on these to make simple statistical machine translation systems in the 'noisy channel' paradigm, how data driven approaches then found ad hoc solutions to many of the challenges of MT during the 2000s and finally how Neural Machine Translation has changed the types of problems we work on today.

Todays homework assignment involves implementing some word alignment models.