Redundant span created from raw text prediction and initial text

[ovbystrova]

Based on the model prediction (string) there is one redundant span created.

Raw text prediction: "Microsoft is an ORG, Microsoft Chine is an ORG, Matt Scott is a PER"

(173, 182, "ORG") -> "Microsoft" (173, 188, "ORG") -> "Microsoft China"

"Microsoft" should not be included in the span (173, 188). It should only be included in (3,12 , "ORG")