Entity Localization Bug: Symbol. pp_t subsymbols are grouped incorrectly in paper 1906.04604v1

[andrewhead]

Assign this issue two labels, one for each of:

1. The entity type (citation or symbols)
2. The localization issue type (missing-entity-detection or bad-entity-detection)

Description: As you can see, the symbol pp_t is grouped into subsymbols as follows:

• one for the first p
• one for the second p and the subscript t (i.e., p_t)
• one for the second p alone
• one for the t subscript

URL (optional): If you run a local instance of the UI that connects to the dev_symbol_failure schema from the database, you can see this behavior at the bottom of page 4 at http://localhost:3001/?file=https://arxiv.org/pdf/1906.04604v1.pdf&preset=demo.

How to fix (optional): The ideal behavior is that:

• The first two ps (pp) are grouped into a single symbol.
• t is made into a subscript for the ps

This will likely require modifying parse_equation.py to, when merging identifiers in a row, also merge in the contents of subscript and supersciprt elements (i.e., 'msub', 'msup', 'msubsup') into the elements before it, if the base element is mergeable (i.e., if it's a letter, and the elements before it in the row are also letters). Test cases should include:

• 'foo_{i}bar_{j}' -> '{foo}_i', '{bar}_j'
  *'22foo_i' -> '22', 'foo_i'