ANNOTATION GUIDELINES

Preliminary comments

Each corpus has its specific set of relationship that are detailed in a separate documents. See DD, DE, FR, GB and US annotation guidelines

General case

The labelling of relationship between entities follow the general rule that the relation should go from the object to its attributes. For example, in the case of a link between an entity LOC and an entity INV, the relationship should go from INV to LOC.

Specific cases

From the set of annotated entities, the context should be sufficient to use the general rule without ambiguity. Standard cases are presented in Examples 1 to 3 below. Two specific cases are worth mentioning:

• Multiple similar objects for a given subject (e.g. two LOC for a same ASG)
• Multiple subjects for a given object (e.g. two INV for a same CIT)

These cases can happen for two reasons, either because the context commands it (see examples 4 and 5) or because one of the entities has been split into multiple parts for example because of a bad OCR, or because of the wording (see example 6).

In any cases, all the corresponding relationship should be annotated.

Examples

Example 1

Example 2

Example 3

Example 4

Example 5

Example 6