The use of automated bilingual term extraction methods for Dutch and French parallel texts in Belgium
Tom Vanallemeersch, Hendrik J. Kockaert · Lirias · 2008
We investigate the use of sentence alignment and bilingual term extraction tools for partially automating the study of phraseological inconsistency in the translation process and making that study more time-effective and comprehensive. We describe the compilation of a large corpus of Belgian legal documents (French arrests translated into Dutch). We apply the alignment tool GMA and the statistical extraction tool TermCalc, which contains a by-catch feature allowing for the detection of alternative translations. We obtain a precision of 87% for the sentence alignment and 82% for the highest-ranking phraseology candidates. The recall, determined by comparing the candidates to a manual study of a Master student at Lessius/KULeuven, is 42%, but TermCalc also adds Dutch equivalents. The low recall is due to the precision-oriented mechanism of the tool and the fact that some phraseologies appear in different forms.