annotátor

An annotátor, or annotator, is a person or software component that adds annotations to raw data. Annotations provide structured information, labels, or metadata that make data interpretable by humans or machines. In data science, annotators create labeled datasets used to train, validate, and evaluate machine learning models, while in linguistics and corpus work they mark linguistic features or annotations on texts.

Annotation occurs across modalities. Text annotation includes tasks such as part-of-speech tagging, lemmatization, named-entity recognition, and sentiment labeling. Image and video annotation covers bounding boxes, segmentation, polygonal outlines, keypoint labeling, and temporal labeling. Audio annotation may involve transcription, speaker labeling, and timestamping. Annotations can be manual, semi-automatic with model-assisted tools, or fully automatic with human review.

The annotation workflow typically involves guidelines, qualification tasks, sample checks, and ongoing quality control. Inter-annotator agreement is often measured to assess consistency, and adjudication may resolve disagreements. Redundancy and sampling help estimate reliability and detect drift.

Common tools support labeling tasks and project management, and many platforms offer web-based interfaces for crowdwork or expert annotators. Examples include Label Studio, Brat, Prodigy, and VGG Image Annotator. Data labeling is essential for supervised learning, NLP corpora creation, medical informatics, and digital humanities projects, among others. Ethical considerations include privacy, consent, bias, and the need for transparent guidelines to ensure data quality and fairness.