Right now maybe_balance_style_tags only does the balancing if the tag is inmediately before the start of the token, giving some whitespace tolerance. However, from running the citation extractor we have seen the following fail cases and possible improvements:

• introductory words included in the style tags: mostly see

  • Example: "<em>See id.</em> at 642"
  • Another: "(see id. at 648-650"
  • Example for reference citation: "<i>see Luperon</i>"

• party names included in the style tags

  • before supra; example : "<i>AT&T, supra</i>" ; "<i>South Seas Yacht Club, supra</i>"

• full case names included in the style tags, when we are only looking for a party name

  • Example of reference "it established in <i>State v. Wingler</i>"

From these examples, I think we could search for the matching style tag for a TOLERANCE number of any character, and include it in the span if we find it. That will help catch case names, and other "introductory" words that are not "see".

This should not create more overlapping issues than already exist; because full span overlaps have already been resolved previous to this annotation step