Cognitive load is a major source of processing difficulties in both interpreting and monolingual speech. This article focuses on measurement of cognitive load by examining the occurrence rate of the disfluency uh(m) in two corpora of naturalistic language: the EPICG, with specific reference to Dutch interpretations of French source texts in the European Parliament; and the sub-corpus of non-interpreted parliamentary speeches from the Spoken Dutch Corpus. In both corpora, the frequency per utterance of uh(m) was studied, in relation to delivery rate, lexical density, presence of numbers and formulaicity (i.e. the number of N-grams), as a Generalised Additive Mixed-effects Model: the frequency of uh(m) in interpretations increases with the lexical density of the source text, while it is inversely related to the formulaicity of both the source text and the target text. These findings indicate the maintenance of a cognitive equilibrium between input load and output load.