• Viterbi Pruning:
  
  
  • takes place at the lowest level after evaluation of the statistical model
    
    
  • traces with the same history can be compared - the best scorer is propagated and the other is deleted
    
    
  • must have efficient storage scheme so that we only have to look at a small number of traces to determine those that are comparable
    
    
• Beam Pruning:
  
  
  • takes place after evaluations and after traversing up
    
    
  • find the best score at a level
    
    
  • keep only traces at that level that fall within a beam of the best
    
    
• Instance Pruning:
  
  
  • takes place after evaluations and after traversing up
    
    
  • traces belong to an instance if they have the same history regardless of timing
    
    
  • keep only a certain number of instances - all others are pruned off and the traces belonging to those instances are also pruned off