Skip to content

Latest commit

 

History

History
48 lines (28 loc) · 1.94 KB

existing_annotation_guidelines.md

File metadata and controls

48 lines (28 loc) · 1.94 KB

Existing Annotation Guidelines

There already exist a few documents with annotation guidelines for entity linking or entity recognition.

TAC

Ellis et al., 2016

Link: https://tac.nist.gov/2016/KBP/guidelines/TAC_KBP_2016_EDL_Guidelines_V1.1.pdf

Ellis et al. consider only entities of the types person, organization, location, facilities and geo-political entities. They do however also include non-named (nominal) mentions of entities of these types.

MAPA

Arranz et al., 2020

Link: http://www.elra.info/media/filer_public/2022/05/10/mapa_annotation-guidelines-v6.pdf

Arranz et al. define entity recognition annotation guidelines for the purpose of anonymizing data. They define three levels of entities. Level 1 consists of the entity types person, time, location, organization, amount and vehicle. Level 2 entities are properties of level 1 entities such as profession, age, marital status or address. Level 3 entities are either components within an entity or types of an entity. These are for example title, family name, street or postcode.

Impresso

Ehrmann et al., 2020

Link: https://zenodo.org/records/3604227

Ehrmann et al. define a set of very thorough annotation guidelines for named entities. The guidelines are derived from the Quaero guidelines for the purpose of extracting information from historical newspaper articles. Non-named entities are not considered. Explanations are in English but examples are given in French or German.

The NewsEye Named Entity and Stance Annotation Guidelines are based on the Impresso guidelines and are similar in large parts. The NewsEye annotation guidelines can be found here.

Quaero

Rosset et al., 2011

Link: http://www.quaero.org/media/files/bibliographie/quaero-guide-annotation-2011.pdf

Quaero defines annotation guidelines for named entity recognition. Sadly, the entire document is in French except for the abstract.