... of the Association for Computational Linguistics, pages 1048–1057,Uppsala, Sweden, 11-16 July 2010.c2010 Association for Computational LinguisticsA Statistical Model for Lost Language DeciphermentBenjamin ... translations, mea-sured with respect to complete word-forms andmorphemes, for the HMM-based substitution ci-pher baseline, our complete model, and our model without the structural sparsity priors. ... generativeBayesian model. This model assumes that eachword in the lost language is composed of mor-phemes which were generated with latent coun-terparts in the known language. We model bilin-gual...