Theoretical Foundations for Digital Text Analysis

Much of social life now takes place online, and records of online social interactions are available for social science research in the form of massive digital text archives. But cultural social science has contributed little to the development of machine‐assisted text analysis methods. As a result few text analysis methods have been developed that link digital text data to theories about culture and discourse. This paper attempts to lay the groundwork for development of such methods by proposing metatheoretical and theoretical foundations suitable for machine‐assisted semantic text analysis. Metatheoretically I draw on the work of Elder‐Vass (2012), Kaidesoja (2013) and others to argue that digital text analysis methods ought to be (and in practice implicitly are) based on a realist constructionist ontology that treats discourses as ontologically real emergent social entities that have causal relationships with non‐discursive social and cognitive processes. Theoretically I follow Feldman (2006) and many others in arguing that language is fundamentally shaped by processes of embodied cognition. Researchers developing digital text analysis techniques must theoretically account for such processes if they wish to produce algorithms that can interpret texts in ways that supplement, and not only amplify, human interpretation. I critically survey contemporary text analysis methods that implicitly share these metatheoretical and theoretical positions and discuss some ways these can be further developed with newly available software.