write rules for NER to capture crowdsourced patterns

[kdutia]

• [x] avoid splitting on ampersand (common with ORG). [45] works fine on jupyter. Have raised spacy issue

• [X] remove 'the year' in the date eg 'the year 1883' or just 'the year' [28, 29]. EntityFilter removes phrase

• [x] misses card index names e.g. 'Foster ,J.M.,' [2] identifies as ORG now which is fine

• [x] pick up firstname last pairs with Ms [A]... [B]... etc. Then mark all firstname/lastname pairs with that same lastname as people. [262] fixed by removing overwrite

• [x] include royal prefixes in names [243] EntityFilter adds prefixes

• [X] capture collections: [A]... collection (mark as ORG) [233] added collection patterns to hc-nlp

• [ ] ~feed list of country abbreviations (USA...) [205]~ can't think of any more sensible abbreviations

• [ ] ~date includes NORP (might be a fix of an existing rule) [co91434]~ can't repeat error in jupyter

• [ ] scientific instruments e.g. Milne seismograph; Ruhmkorff coil. Pick up [A]... instrument from list of instruments? [co134618, co54032]

• [x] extend to from 2 to 3 names if no punctuation. Joseph Henry -> Joseph Henry Morton [co8055319] caused by overwrite - disabled

• [x] remove 'n years' from date entities

• [x] test new NER performance

[kdutia]

up another 0.6% F1 from the previous best augmented method: p 64.6, r 72.9, f1 68.5