Neutral subset: contains 904 ambiguous mentions (names of entity spaces) placed into artificially constructed context, designed in such a way that disambiguation is impossible (no useful information can be derived from the context). Every mention is substituted into 7 "neutral" templates to create such context. The subset was used for obtaining the prior-based predictions of the ED models
Tail subset: contains 904 entities controlled for low commonness but not for ambiguity (mostly unambiguous). Not linked to the same entity spaces as the Top and Shadow subsets; serves as a control group. Every entity is provided with a short context example, which was obtained from a web page specified as an external reference of a Wikipedia entry about the corresponding entity
Shadow subset: contains a shadow entity (one whose commonness score is lower than that of the Top entity) for each of the 904 entity spaces. Every entity is provided with a short context example, which was obtained from a web page specified as an external reference of a Wikipedia entry about the corresponding entity
Top subset: contains the top entity (one with the highest commonness score) for each of the 904 entity spaces. Every entity is provided with a short context example, which was obtained from a web page specified as an external reference of a Wikipedia entry about the corresponding entity
