Measuring Translation Literality

Tirkkonen-Condit (2005: 407–408) argues that “It looks as if literal translation is [the result of] a default rendering procedure”. As a corollary, more literal translations should be easier to process, and less literal ones should be associated with more cognitive effort. In order to assess this hypothesis, we operationalize translation literality as 1. the word-order similarity of the source and the target text and 2. the number of possible different translation renderings. We develop a literality metric and apply it on a set of manually word and sentence aligned alternative translations. Drawing on the monitor hypothesis (Tirkkonen-Condit 2005) and a model of shared syntax (Hartsuiker et al. 2004) we develop a model of translation effort based on priming strength: shared combinatorial nodes and meaning representations are activated through automatized bilingual priming processes where more strongly activated nodes lead to less effortful translation production. The theoretical framework explains the observed production- and reading times and justifies our literality metric.