Table 2.

IR Results on epic-qa-dev regarding the fields used as a query: (i) query; (ii) query+question: query and question concatenated; and (iii) $w$(qry+qs)+($1-w$)backg: complex query built concatenating query and question fields, and combining linearly the concatenation with the background field.

(a) IR Results on epic-qa-dev for passage retrieval over 5000 element rankings with different query building strategies.
Query building	NDCG	R@500	R@1K	R@2K	R@3K	R@4K	R@5K
query	0.2977	0.2864	0.3766	0.4606	0.5397	0.5689	0.5918
query+question	0.3832	0.3906	0.4898	0.5848	0.6599	0.7101	0.7346
0.5 $\cdot$(qry+qs) + 0.5 $\cdot$backg	0.3901	0.3979	0.5126	0.6099	0.6795	0.7154	0.7604

(b) IR Results on epic-qa-dev for document retrieval over 5000 element rankings with different query building strategies.
Query building	NDCG	R@500	R@1K	R@2K	R@3K	R@4K	R@5K
query	0.4758	0.5548	0.6518	0.6951	0.7308	0.7494	0.7602
query+question	0.5555	0.6478	0.7492	0.8102	0.8382	0.8566	0.8606
0.7 $\cdot$(qry+qs) + 0.3 $\cdot$backg	0.5613	0.6636	0.7575	0.8222	0.8437	0.8589	0.8625