Corpora

Corpora, plural of corpus, are structured collections of authentic language data assembled for linguistic analysis. They may include written texts, transcripts of spoken language, or both, and are typically annotated with metadata such as genre, date, author, and source. Corpora serve as empirical bases for description, variation study, and the development of language technologies.

Corpora vary by content and purpose. General corpora aim to represent broad language use, while specialized corpora focus on particular domains (medicine, law, news). Learner corpora contain data from language learners and are used to study acquisition and error patterns. Parallel corpora align texts and translations across languages for machine translation and cross-lingual research. Spoken corpora contain transcribed speech and may include phonetic or prosodic annotations. Multilingual and diachronic corpora support cross-language comparisons and historical change.

Most corpora are annotated to facilitate analysis, with layers such as tokens, lemmas, part-of-speech tags, syntactic structures, semantic roles, and discourse information. Formats and standards (for example, TEI-XML, JSON, or XML-based schemes) help interoperability, while corpus management systems enable searching, concordancing, and statistics.

Creating a corpus involves data collection, cleaning, annotation, and validation, along with consideration of licensing, representativeness, and bias. Size is often reported in tokens or word forms; statistical power depends on sampling and annotation quality. Limitations include non-representativeness, genre imbalance, and the potential for annotation errors.

Applications include linguistic description, lexicography, language education, and the development and evaluation of natural language processing tools such as tokenizers, parsers, and language models. Corpora remain central to empirical approaches in linguistics and language technologies.