Data from natural languages (in contrast to, say, the results of psycholinguistic experiments) are still a major source of evidence used in linguistics, whether they are elicited through grammatical judgments, as in generative linguistics, or by collecting samples, as preferred in typology. The underlying assumption is that data are alike in their value as evidence if they occur in natural languages. The present paper questions this assumption in showing that there is a difference in the naturalness of languages because languages like German or English have originally emerged as secondarily learned written languages, that is they once were languages without native speakers. Although they are nowadays acquired as first languages, their grammars still contain inconsistent properties which partly disqualify standard languages as a source of evidence.
Mid-level generalizations of generative linguistics: Experimental robustness, cognitive underpinnings and the interdisciplinarity paradox