This chapter describes an empirical study to examine the concept of variation in multiple TTs of the same ST and the subjectivity that this may entail. Although the main object of study are translation products (i.e. the TTs themselves), it is motivated by a keen interest in the translation process, in the differences that exist between translators, and, the other side of the same coin, in what characteristics translators share and how this can be traced in their work. The investigation centres on the small linguistic variations between different target versions of the same source text and on the extent to which those elements that are most open to shifting are expressions of attitude and feelings. This will shed further light on the question of ‘legitimate translation variation’, which is ‘the fact that independent human translations of the same text often use different words and structures to convey the same content’ (Babych and Hartley 2004: 833). In the context of statistical calculation of the quality of machine translation output measured against TTs produced by human translators, Babych and Hartley analyse two human translations of a French legal text into English. They fi nd that the stability of units across the human translations is related to the salience of those units in the text:

Highly signifi cant words, which are consistently used within a single translation, were found to be the most unstable across different translations. The possible reason for this fact could be that translation of signifi cant units typically requires invention of some novel translation strategy.