Paper: Phrase Translation Probabilities with ITG Priors and Smoothing as Learning Objective

ACL ID D08-1066
Title Phrase Translation Probabilities with ITG Priors and Smoothing as Learning Objective
Venue Conference on Empirical Methods in Natural Language Processing
Session Main Conference
Year 2008
Authors

The conditional phrase translation probabil- ities constitute the principal components of phrase-based machine translation systems. These probabilities are estimated using a heuristic method that does not seem to opti- mize any reasonable objective function of the word-aligned, parallel training corpus. Ear- lier efforts on devising a better understood estimator either do not scale to reasonably sized training data, or lead to deteriorating performance. In this paper we explore a new approach based on three ingredients (1) A generative model with a prior over latent segmentations derived from Inversion Trans- duction Grammar (ITG), (2) A phrase ta- ble containing all phrase pairs without length limit, and (3) Smoothing as learning ob- jective using a novel Maximum-A-Posteriori version of D...