corpusdriven

Corpusdriven, often written as corpus-driven, is a term in linguistics describing approaches in which evidence from large text corpora guides linguistic description and theory. In corpusdriven research, patterns and generalizations are expected to emerge from actual usage, with hypotheses revised to fit observed frequencies, collocations, and constructions, rather than being imposed from preconceived categories.

Methodology and scope: Researchers analyze large corpora—written, spoken, or mixed—using statistical and computational tools to uncover frequency patterns, collocations and multiword expressions, syntactic constructions, and variation across registers or genres. Outputs include distributional profiles and evidence of semantic prosody, which reflect evaluative associations in context. Data sources range from balanced general corpora to domain-specific or web-derived corpora.

Relation to theory and limitations: Some scholars treat corpusdriven work as distinct from corpus-based studies, emphasizing discovery over verification; in practice the distinction is blurred. Benefits include empirical grounding, discovery of usage patterns beyond intuition, and relevance for lexicography and language teaching. Limitations include corpus representativeness, annotation biases, and the risk of drawing broad conclusions from limited data; results are typically interpreted alongside qualitative analysis and existing theory.