In addition, the final question in each series is an explicit “Other” question, which is to be interpreted as “Tell me other interesting things about this target I don’t know enough to ask directly”. This last question is roughly equivalent to the definition questions in the TREC 2003 task.
The Other questions were evaluated using the methodology originally developed for the TREC 2003 definition questions. A system’s response for an Other question consisted of an unordered set of [doc-id, answer-string] pairs.
AI came from NIST TREC QA. Please stop this buzz and study at least something. Otherwise you all look like idiots.