Statistical Machine Translation through Global Lexical Selection and Sentence Reconstruction
Srinivas Bangalore, Stephan Kanthak, Patrick Haffner · The MIT Press eBooks · 2008
This chapter presents the details of training a global lexical selection model using classification techniques and sentence reconstruction models using permutation automata. It also provides a stochastic finite-state transducer (SFST) as an example of an approach that relies on local associations, and uses it to compare and contrast the approach. The chapter is organized as follows. Section 10.2 describes in detail the different stages used to train an SFST translation model, and discusses the steps in decoding a source input using the trained SFST model. Section 10.3 presents the global lexical selection and the sentence reconstruction models. Section 10.4 discusses the rationale for choosing the classifier to train the global lexical selection model. Section 10.5 reports the results of the two translation models on a few data sets and contrasts the strengths and limitations of the two approaches.