• support for arbitrary grammars, N-grams and dynamic LMs (combinations of grammars and N-grams).
  
  
• Lexical tree organization of lexicon
  
  
  
• Word identity unknown at intermediate nodes - let traces hold the LM history
  
  
• Viterbi approximation used for word internal nodes
  
  • use max LM score for word internal nodes
  • apply real LM score at word ends