The study of literature abounds in theories. Since the time of Plato, scholars have developed theories about the nature and function of literature. And yet until recently, we have had no way of empirically testing these theories, nor developing new theories based on large-scale observations of literary texts. In this talk I will present research that aims to motivate three emerging hypotheses about the social function of fictional storytelling based on the use of computational approaches to the study of literature. These hypotheses thus represent the first steps in a data-driven theory of literature. I call them the coherence hypothesis, immutability hypothesis and the phenomenological hypothesis. Coherence refers to the degree of semantic and stylistic distinctiveness of fictional versus non-fictional discourse. Immutability refers to the transhistorical (and potentially transcultural) continuity of such distinctiveness. And phenomenological refers to fiction's unique investment in the fictional subject's sensing, testing and wondering relationship to the world. The talk will explore the appropriateness of different kinds of data for such analysis, the choice and interpretation of different feature spaces, as well as the available methodologies that can be used to support the development of such hypotheses.