EXAM PREPARATION

Machine Translation 1.


* Provide an illustration for the following: Translation is made more
  difficult by the fact that

  - The translator must add or subtract substantive information from the
    message that the text conveys.
  - The meanings of many vocabulary items are bound to the culture of the source text and thus have no correspondents in the target language.

* In a rule-based machine-translation system, what role would be played by
  - a morphographemic transducer 
  - a morphological analyzer/generator
  - syntactic parser/generator
  - a transfer component

* Bernard Vauquois used a triangular diagram to discuss various
  strategies used in machine translation. The following question
  relate to such a triangle.

  - Draw the diagram
  - Why is the diagram narrower at the top?
  - What is the top point in the triangle called and what is its role in machine translation?
  - What would the diagram look like for a syntax-based transfer (or some other) system?
  - In what sense are higher levels in the triangle more "abstract"?     

* Explain briefly:   
  - Triangulation
  - Interlingua
  - Vauquois' triangle
  - Robustness
  - Transfer 
  - Reference

* Why might the following be difficult to translate into another
  language, and why?
  - goal keeper
  - alimony
  - nut
  - hot tea
  - the bill passed the house
  - value-added tax  
  - the world cup

* Why would a language with one of the following properties be
  easier/harder for a rule-based/statistical system to translate
  into/out of English?
                                                                     
  - Rich morphology
  - Genders
  - No determiners
  - Noun-noun compounds
  - VSO, SOV, free, word order
  - Separable verbs or verbal particles (like German "ankommen" or "to look up"

* The following properties are sometimes claimed as advantages of
  statistical/rule-based machine-translation. Which are generally seen
  as favoring which side, and why? 
  - (Non)determinsim
  - Rapid development
  - Broad domain coverage
  - Robustness
  - Ability to handle long-distance dependencies. 
  - speed 
  - better resolution of ambiguities
  - Fluent translation 
  - Accurate translation 

* Advocates of statistical machine translation have criticized earlier
  rule-based approaches to the problem an a variety of grounds.
  Defend them briefly against the following specific criticisms:

  - linguists are only interested in exotic or fringe phenomena.
  - the time and effort required to write a useful set of rules makes them 
    impracticable.
  - linguists recognize ambiguities but provide no help in resolving them.  
  - The rule that linguists write do not constitute robust systems.      

* Explain briefly the main import of Zipf's law and explain its
  relevance to the construction of linguistic models by machine
  learning based on large corpora.