Natural Language Hypotheses in Scientific Papers and How to Tame Them

Abstract

AbstractHypotheses are critical components of scientific argumentation. Knowing established hypotheses is often a prerequisite for following and contributing to scientific arguments in a research field. In scientific publications, hypotheses are usually presented for specific empirical settings, whereas the related general claim is assumed to be known. Prerequisites for developing argumentation machines for assisting scientific workflows are to account for domain-specific concepts needed to understand established hypotheses, to clarify the relationships between specific hypotheses and general claims, and to take steps towards formalization. Here, we develop a framework for formalizing hypotheses in the research field of invasion biology. We suggest conceiving hypotheses as consisting of three basic building blocks: a subject, an object, and a hypothesized relationship between them. We show how the subject-object-relation pattern can be applied to well-known hypotheses in invasion biology and demonstrate that the contained concepts are quite diverse, mirroring the complexity of the research field. We suggest a stepwise approach for modeling them to be machine-understandable using semantic web ontologies. We use the SuperPattern Ontology to categorize hypothesized relationships. Further, we recommend treating every hypothesis as part of a hierarchical system with ‘parents’ and ‘children’. There are three ways of moving from a higher to a lower level in the hierarchy: (i) specification, (ii) decomposition, and (iii) operationalization. Specification involves exchanging subjects or objects. Decomposition means zooming in and making explicit assumptions about underlying (causal) relationships. Finally, operationalizing a hypothesis means providing concrete descriptions of what will be empirically tested.