Errors of Omission in Translation

Automatic detection of translation errors represents one of the more promising applications of NLP techniques to this domain. This paper concentrates on one class of error, the inadvertent omission. To a greater extent than ‘false friends’, terminological inconsistency, etc., the detection of omissions raises problems both theoretical and practical in nature. These problems are discussed, and a technique is presented for identifying possible omissions in a completed translation by employing a model of translational equivalence between words. Examples are taken from a varied corpus of French-English bitext, and illustrate how different settings of the parameters of the system affect its performance. The approach is implemented as part of a translation-checking program.