PROBABILITY MODEL OF UNION BASED ON PHRASES FOR STATISTICAL AUTOMATIC TRANSLATION.

摘要

Procedure implemented by computer to generate a joint probability model based on phrases from a parallel body comprising a plurality of sentences in the source language and a corresponding plurality of sentences in the target language; the procedure comprising: a) defining from the parallel body high frequency n-grams ( vec (ei) in E, and ( vec (fi) in F, where E and F comprise phrases in the source and target language , respectively; b) obtain an initial joint probability distribution t based on the sentence, by: i) taking, for each pair of sentences (E, F) in the body, three Cartesian products of n-grams ( vec (ei ) in E, and ( vec (fi) in F; ii) determine, for each pair of n-grams (ei, fi) in the Cartesian product, a count t given by the expression: ** (See formula) * * where l and m are the lengths of the phrases E and F, respectively, a and b are the lengths of the n-grams ( vec (ei) and ( vec (fi), and S is the second class Stirling number; iii ) add the accounts t and normalize, and c) perform the Maximum Expectation training for a plurality of iterations to generate a joint probability distribution t.