What Might Corpora Tell us about Language?

This chapter establishes a theoretical framework for corpus linguistics that challenges the ‘bigger is best’ approach, advocating for theory-rich linguistics supported by richly annotated corpora. The author proposes a sophisticated methodological framework centred on the ‘3A perspective’ (Annotation, Abstraction, and Analysis) for principled corpus research. Three distinct classes of empirical evidence are identified: factual evidence (existence of linguistic events), frequency evidence (distributional patterns), and interaction evidence (probabilistic relationships between phenomena). A crucial distinction is drawn between exposure rates (frequency per million words) and choice rates (probability of selecting alternatives), with choice rates providing more theoretically meaningful insights. Two case studies demonstrate the perspective: an investigation of modal auxiliary shall versus will decline using DCPSE reveals systematic change when framed as speaker choice rather than exposure frequency; and analysis of noun phrase premodification in ICE-GB uncovers systematic negative feedback in speakers’ decisions to add sequential adjectives. The chapter distinguishes framing constraints from interaction evidence, separating grammatical rules that permit/prevent choices from genuine interactions between independent decisions. Rich annotation serves as a ‘handle on the data’ enabling reliable abstraction regardless of theoretical disagreements. The framework positions corpus linguistics as an observational science generating novel empirical findings requiring theoretical explanation.