Decoding for Word Based Models
When decoding was first encountered in word-based models, a quite diverse set of decoding algorithms was explored.
Word Based Decoding is the main subject of 5 publications.
Several decoding methods for word-based models are compared by Germann et al. (2001)
, who introduce a greedy search (Germann, 2003)
and integer programming search method. Search errors of the greedy decoder may be reduces by a better initialization, for instance using an example-based machine translation system for seeding the search (Paul et al., 2004)
. A decoding algorithm based on alternately optimizing alignment (given translation) and translation (given alignment) is proposed by Udupa et al. (2004)