The identification of abstractions, i.e. terms that have a particular significance in a given domain, and such that they can indirectly characterize the most salient features of the document in which they appear, has often been recognized as a useful tool in the analysis of domain descriptions and requirements documents in software development. In this paper we propose a new technique for the identification of single- and multi-word abstractions named Relevance driven abstraction identification (RAI) and a corresponding tool implementation, present an experiment comparing the effectiveness of our technique with human judgement and with a different technique proposed in the literature, and discuss a number of ways in which the abstractions so identified can be used to good profit in requirements engineering.