Adapt ConText algorithm with modifications as in ContextD

[lcreteig]

See the "ContextD: ConText for Dutch" section in the ContextD paper

• GP specific rules.

  [...] general practitioners often negate the existence of a clinical condition by putting a minus sign after the term, e.g., fever-. We added a couple of rules to catch such occurrences (and their variations) of negation in the GP text.

• Combined triggers. Some triggers have both a forward (before the entity) and backward part (after the entity). For example "nooit trauma gehad". Apparently performance increases if you only look for nooit if gehad occurs as well (so you don't catch things as "nooit bloedverdunners gebruikt"). There's no direct support for this in medspacy, but you can add a post processing rule in case "nooit" was detected. See this issue for an example.
• Scope of trigger terms. Normally this is until the end/beginning of the sentence. But this was modified for specific document types:
  • 6 words for GP entries (still terminated by termination triggers, as well as "some punctuation, such as comma and semicolon"
  • 10 words for specialist letters (still terminated by termination triggers; "For specialist letters, only colon and semicolon were added to the termination triggers")