Engaging with bad (meta)data in historical corpus linguistics

In this chapter, we discuss some common pitfalls related to historical data and its use in linguistic analysis. We argue that the “philologist’s dilemma”, as originally proposed by Rissanen (1989), should be reconceptualized to meet the needs of the fast-evolving field of corpus linguistics, where scholars make increasing use of big-data resources and sophisticated statistical modelling. By providing examples of errors and uncertainties related to, for example, corpus metadata, sampling, balance, and OCR accuracy, we argue that corpus linguists should pay increasingly close attention to the sampling and annotation principles employed in the compilation of historical corpora as well as to the quality of the linguistic data. We propose that the principle of “knowing one’s corpus” in terms of its compilation principles has become all the more important in the age of big-data corpora, where it is not feasible for individual researchers, or corpus compilers, to validate their data manually.