Formalization and Evaluation of Candidate Translation Ranking
Estelle Maryline Delpech · 2014
This chapter describes the exploratory work meant to experiment on several ranking methods for the candidate translations. It starts by detailing the chosen ranking criteria, and then indicates the way in which the criteria can be combined. The results are given, and the limits of the work as well as the research prospects are also discussed. Ranking criteria all have a value between 0 and 1, but they present very different value scales. The ranking methods are evaluated on the data set E, which corresponds to the source terms of the a priori reference for which the system generated a translation. The training data correspond to the data set T, i.e. the source terms, which do not belong to the a priori reference and for which the system generated a translation. This chapter concludes with a general discussion of the results obtained.