This study aims at developing a method for investigating the invariants across different scientific discourses. The paper illustrates a model analysis for processing general scientific vocabulary by applying three complementary approaches: cross-disciplinary, onomasiological and phraseological. The combination of these approaches allows us to bypass traditional methods generally based on a lexical analysis of a single domain. We first define the borderline between the specialised scientific language features and those features that are common to various scientific discourses, then we present a corpus-based study of a combinatorial profile of lexical resources used for formulating general scientific ideas. This method thus provides a framework for the modelling of general scientific language resources.