Lexical postprocessing by heuristic search and automatic determination of the edit costs

We describe the realization of a dictionary based lexical postprocessing approach. A character hypotheses lattice (CHL) serves as input which is compared with the words of the vocabulary, using a generalization of the weighted edit distance. The search for the best word is based on a depth first traversal through the paths of the CHL and is directed by several heuristics to achieve a reasonable processing speed without deteriorating the recognition rate significantly. An iterative supervised automatic learning algorithm is proposed which determines the costs for the edit operations. Experiments reveal that this method significantly improves the recognition accuracy.