EXAMPLE: STRING SIMILARITY

• Consider the problem of measuring the distance between two strings below:
  
  
  
• Transition Penalty: Any non-diagonal transition has a penalty of 1 unit.
  
  
• Node Penalty: Any two dissimilar letters that are matched at a node incur a penalty of 1.
  
  
• How much memory do we need if we just want to keep the best score?
  
  
• When would we want to retain the backpointers at each node?
  
  
• How can we constrain the search space to make this algorithm more efficient for speech processing?