Experimental Data
Estelle Maryline Delpech · 2014
This chapter presents the data used to test the translation method: comparable corpora, source terms, reference lexica, training data and linguistic resources. The methods are applied for the translation from English to French and English to German. First, the chapter discusses the comparable corpora used, then the source terms selected to test the translation method as well as the data used for the evaluation of the translation generation and the translation ranking training and evaluation datasets. Finally, it describes the linguistic resources on which the translation system relies. To build a priori reference, the chapter uses the UMLS (Unified Medical Language System), a medical domain metathesaurus. This metathesaurus is often used as a reference for the machine translation of terms, especially when evaluating data-driven methods since it also provides training examples.