A Tree-Trellis N-Best Algorithm for Decoding in Maximum Entropy Models
Feng Chon · 2005
Maximum entropy models have been widely adapted in various natural language processing tasks. But there are some deficiencies in the decoding algorithm used by many previous researches. A n-best tree trellis algorithm is pro- posed for decoding in maximum entropy models. The performance analysis and comparison with other decoding algo- rithms are also presented. Another advantage of our method is that the possible collision in action sequences can be de- tected and eliminated.