corpusAnnotation

Corpus annotation is the process of adding linguistic information to a text corpus to enable systematic analysis and computational processing. Annotated corpora provide structured data that supports linguistic research, language technology development, and the evaluation of natural language processing systems.

Annotation layers may include lexical items (tokenization, lemmatization), morpho-syntactic information (part-of-speech tags, morphology), syntactic structure (parsing trees or dependency relations), semantic roles, named entities, coreference, discourse relations, sentiment, and pragmatic cues. Each layer is governed by a specific annotation scheme with defined labels and guidelines to ensure consistency.

Annotation schemes aim for reliability across annotators. The process typically involves training, multiple annotators working in parallel, quality checks, and adjudication to resolve disagreements. Inter-annotator agreement measures, such as Cohen’s kappa or Fleiss’ kappa, are used to assess reliability and guide guideline refinements.

Common data formats and tools foster interoperability. Formats include CONLL-U for morphosyntactic annotation, CoNLL-derived formats, TEI/XML, and JSON or XML for multi-layer data. Widely used annotation tools include BRAT, WebAnno, GATE, and INCEpTION, among others, often supporting export to standard formats for research and model training.

Applications span linguistic research, corpus linguistics, language technology development, and machine learning, where annotated data serve as training, evaluation, or benchmark resources. Challenges include subjectivity in labeling, cross-linguistic variation, aligning multiple annotation layers, data sparsity for less-studied languages, and ensuring reproducibility and clear licensing for shared corpora.