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