Märgistusega

Märgistusega is the Estonian term used to describe the act or result of tagging, labeling or annotating data, text, images or other content. It is formed from the verb märgistama (to mark or tag) with an affix that conveys the idea of “with tagging/annotation,” and is commonly used in information science, linguistics and data work to refer to annotated material.

In practice, märgistusega refers to the process of adding metadata through agreed labels or tags that describe content, category, or properties. In supervised machine learning, datasets are created as märgistusega data, where each item carries one or more labels that identify its target class. In natural language processing, texts may be märgistusega with parts of speech, named entities, sentiment or argumentative structure. In computer vision, images or videos are marked with bounding boxes, segmentations or action labels. The concept also appears in digital humanities and archival work, where documents are annotated to indicate topics, provenance or linguistic features.

Methods for producing märgistusega data range from manual annotation by trained specialists to semi-automatic approaches that propose labels for human validation, and to crowdsourcing when large-scale labeling is needed. Ensuring quality involves clear annotation guidelines, measuring inter-annotator agreement, and iterative revisions to improve consistency and reduce bias.

The availability of high-quality märgistusega data is central to training and evaluating computational models, enabling reproducibility and transfer across tasks and domains. See also annotation, labeling, metadata, dataset.