The application of corpus methodology to translation: the JPED parallel corpus and the Pediatrics comparable corpus

Abstract

This Masters dissertation describes the process of designing and compiling a suite of three corpora and the software solutions developed to access them. It then goes on to investigate the general statistical characteristics of the three corpora and finally an analysis is performed of a specific language feature across all three corpora. The three corpora are all made up of paediatric texts. The first corpus consists of 785,448 words, comprising the original Portuguese texts of all content published online by JPED in 2003 and 2004. The second corpus consists of the English translations of the texts in the first corpus. It contains 771,169 words and has been aligned with the first at the paragraph level. The third corpus consists of 772,090 words of original English texts published online in Pediatrics from January to May 2004. The third corpora is designed to be comparable with the first two. The general statistical characteristics of the three corpora suggest that English translations of Portuguese paediatric texts have more in common with paediatric texts originally written in English than with Portuguese paediatric texts, suggesting that the language (English or Portuguese) is dominant over translation status. The choice of whether to use an adjective or a noun when translating four Portuguese adjectives with classical etymology was investigated using the three corpora described. It was observed that, in the JPED translations, the frequency of usage use of "cardiac" or "heart" and "hepatic" or "liver" were approximately equal while the frequency of usage of "pulmonary" or "lung" and "renal" or "kidney" were roughly one order of magnitude different. These distributions approximate to binary choice systems described by Halliday as "equi" (50:.50) and "skew" (90:10). In the Pediatrics corpus, cardiac/heart and pulmonary/lung were "equi" systems while hepatic/liver ( 10:90) and renal/kidney ( 90:10) were "skew" systems.