Either just add attested English usages for c, d, and `i, or import corpus examples with those annotations

[nschneid]

Importing the corpus examples would require heuristically figuring them out from STREUSLE lexcat + POS information (because STREUSLE doesn't preserve their original use by annotators).

[nschneid]

(It would be easy to add `i because these correspond to lexcat INF. Is it worth having them or would they get in the way of browsing actual supersense annotations for TO and FOR?)

[nschneid]

`c

A tquery.py search in STREUSLE shows that `c is easy to recover: simply look for lexcat=CC with an ADP in the lexical expression:

• Single-word ADP

./tquery.py streusle.json lc=CC "ll!=[ ]" upos=ADP
reviews-159371-0002	me and my dad where in NJ for a kc chiefs ( my home team ) >> vs << the NY jets and for game 4 of the world series .
reviews-303922-0004	They are honest about ' immediate ' concerns >> versus << ' recommended ' repairs and have very fair prices .

• Multiword ADP

./tquery.py streusle.json lc=CC "ll=[ ]" upos=ADP
reviews-047184-0005	We have attended A Ward Dance Centre for over a year and really enjoy the friendly and welcoming way we are taught Ballroom and Latin >> as well as << the fun filled social dance evening held every Saturday evening ....
reviews-235423-0012	I do nt know how it is possible to make orange chicken , sesame chicken and kung pao chicken >> as well as << cheese puffs taste THAT bad but China Delight accomplished that .
reviews-310018-0002	She has a great way with children and is able to capture their personality >> as well as << many spontaneous images .
reviews-335815-0002	There are a couple decent people working there , but the rest are VERY dishonest , >> as well as << rude , I have yet to hear the truth come out of their mouths .
reviews-376320-0008	I completely enjoyed my whole check in experience and was impressed with the friendliness and professionalism of the staff >> as well as << the accommodations themselves .

• Multiword non-ADP: these should not be imported

reviews-079375-0009	I live nearly two hours away >> and yet << I will still make the drive to see them !
reviews-376320-0004	This can tend to be a stressful experience in itself >> let alone << adding crossing boarders for the first time .

`d

For lexcat=DISC we have:

• Single-word ADP, ADV, or SCONJ: we need a lexical whitelist of ADVs considered prepositions to exclude "next" and "too"

./tquery.py streusle.json lc=DISC "ll!=[ ]" "+upos=(ADP|ADV|SCONJ)"
reviews-004574-0006	('ADP',)	Secondly , the enchladas did not come with enchilada sauce .. but chili ... >> like << Hormel 's chili .. the cheese was American cheese !
reviews-004940-0004	('ADP',)	>> On << one order , we needed it rushed and shipped to a different state .
reviews-053248-0003	('ADP',)	The truth is , >> in << my and my dining partners ' experience , this is a fine little restaurant with some unique food .
reviews-109263-0003	('ADP',)	I had sonic in many other palces but >> for << some reason this sonic is always just covered in grease and not good ... :(
reviews-117115-0006	('ADV',)	>> Next << , they tried to force the window with a pry bar and then to break the window with a hammer .
reviews-185045-0001	('ADV',)	Rcommended by bees , >> too << !
reviews-228731-0001	('ADV',)	prepared the road test with a driving ... prepared the road test with a driving school in edmonton , but my instructor only trained me in a narrow street , >> hence << I took one 90 minute lesson from the Noble driving school to learn the skill of changing lane , and found them very friendly and professional .
reviews-268673-0005	('ADV',)	>> Besides << parking is a pain .. cramped and un-ruly with Kumon Parents next door .... gives me heebee gee bees'
reviews-294081-0006	('ADV',)	>> NEXT << , THERE SHOULD ONLY BE ONE PERSON BRINGING YOU YOUR FOOD .
reviews-333243-0014	('ADP',)	>> In << the words of my new accountant , THEY LET ME DOWN !
reviews-351561-0018	('ADP',)	17th , >> like << over by 16th and 15th YES , I say , one mile west of you .
reviews-351561-0021	('ADP',)	She says , Is that 17th >> like << over past broad .
reviews-366946-0004	('ADV',)	Filled up on too much beer and >> hence << can not comment on the food .
reviews-385281-0002	('SCONJ',)	I 'm 22 , and my hairdresser was great ( and not " old " >> like << one of the reviews says ) - she really listened to what I wanted and gave me tons of tips on how to style my hair so I could get it to look the way I wanted it to .
reviews-388799-0006	('ADP',)	I will sum it up >> with << , it was worth every penny !

• Other single-word DISC: these are mostly NOUN ("thanks", "congratulations", "PS")

• Multiword starting with ADP, ADV, or SCONJ: again, we need a whitelist to filter the ADVs to remove "short of that", "more often than not", "never mind", .... "Not to mention", for which "not" is (erroneously?) tagged as ADV, is borderline: it's an infinitive verb with negation.

./tquery.py streusle.json lc=DISC "ll=[ ]" "+upos=(ADP|ADV|SCONJ)" | egrep "\('(ADP|ADV|SCONJ)', " | wc -l
      72

• Multiword starting with PART: this includes infinitive "to" ("to boot", "to be clear") and negation ("not a _ to be found"). We should filter out the negations.

./tquery.py streusle.json lc=DISC "ll=[ ]" +upos | egrep "\('PART', " | wc -l
       7

• Multiword not starting with ADP, ADV, SCONJ, or PART: 25 are "thank you"; another 23 are mostly NPs or VPs, and 4 of these are borderline candidates for adpositions:

reviews-196219-0002     ('ADJ', 'ADP')  IT IS NOT A HIGH END STEAK HOUSE , >> MORE OF << THE CUISINE BRETT ENJOYS IN MISSISSIPPI .
reviews-290594-0008     ('VERB', 'ADP') It sounds ( >> according to << your own statement ) that they had a roomful of dogs , so they must be doing something right - and are keeping those dogs safe from potential problems .
reviews-291046-0004     ('VERB', 'ADP') >> According to << news accounts , the company is struggling .
reviews-364454-0005     ('VERB', 'ADP') >> Based on << the test results , the home inspector stated that the quality of their job was “ excellent ” .

[nschneid]

Summary of the above: we could consider importing

• all lexcat=INF tokens as `i,
• lexcat=CC tokens as `c provided that the first word is tagged as ADP,
• lexcat=DISC tokens as `d provided that the first word is one of the following:
  • ADP
  • SCONJ
  • ADV, and lemma found in a whitelist of prepositions
  • PART, and lemma is "to"