An investigation was undertaken into the possibility of automatically detecting how concepts exist in relationship to each other in abstracts, a text-type commonly used in free-text retrieval. The end goal of this research is to capture these relationships in structured representations of abstracts' contents so that users can require not only that the concepts of interest to them co-occur in the retrieved documents, but also that the roles they play in relation to each other are the ones of interest. Four tasks found useful in revealing other schema were performed by expert abstractors. The results were analyzed and used as the basis of developing a frame-like structure of abstracts reporting on empirical work. A discourse linguistic analysis of a sample of 276 abstracts identified the lexical/syntactic clues which could be used by a system to automatically instantiate the frame-like structure of individual abstracts.