One of the crucial challenges of statistical machine translation is the lexical consistency of manually translated words and multiword expressions (MWEs) with multiple occurrences in the source language. In this paper, we present the degree of translation inconsistency and we introduce the index of translation completeness of fixed MWEs. The research was based on the recently developed system that intends to extract the entire candidate MWEs from Orwell’s 1984 parallel corpora and to predict their translations between English, Macedonian, and Slovene.