Subcorpora

Subcorpora are subsets of larger corpora selected for specific research questions or criteria. By focusing on a defined portion of data, subcorpora allow researchers to study language use within particular genres, time periods, registers, or linguistic features without the noise of the full corpus.

Subcorpora are created by applying filters or queries to the parent corpus. Selection can be based on metadata such as genre, author, region, or date; linguistic criteria such as part of speech or lemma; or content criteria such as presence of specific keywords. Subcorpora may be formed through deliberate sampling (random, stratified, or balanced) to ensure comparability across groups or conditions.

Common uses include estimating frequencies and collocations within a domain, comparing linguistic patterns across genres or time periods, and building frequency lists or lexical profiles. They are also used to study diachronic change, register variation, or terminology in specialized fields.

Practitioners must consider representativeness, size, metadata quality, and the potential for selection bias. Subcorpora should be comparable if used in cross-corpus analyses, and documentation should report criteria, size, and sampling method. Many corpus tools, such as Sketch Engine, AntConc, and COCA’s built-in subcorpora, support subcorpus creation and analysis.