UniversalDependencies / UD_English-EWT

English data
Creative Commons Attribution Share Alike 4.0 International
199 stars 42 forks source link

POS tags for 'whatever' #393

Closed amir-zeldes closed 10 months ago

amir-zeldes commented 1 year ago

It seems "whatever" has 4 POS tags in EWT:

The wrong WDT cases should be corrected (I see some of these in GUM and ON as well). I'm not sure if we want UH - OntoNotes seems to treat these literally so GUM does as well. Should we use UH for the exclamation? (considering we don't use UH for a dejected 'what?')

AngledLuffa commented 1 year ago

I'm not sure if we want UH - OntoNotes seems to treat these literally so GUM does as well. Should we use UH for the exclamation?

Whatever...

WDT (mostly used as a real determiner: "whatever X", but some errors that should probably be WP?)

An example of the error you have in mind is this, where it's the head of the noun phrase?

# sent_id = weblog-blogspot.com_dakbangla_20050311135387_ENG_20050311_135387-0217
# text = Whatever your political persuasion, the FBI and CIA deserve our support.
1       Whatever        whatever        DET     WDT     PronType=Int    10      advcl   10:advcl        _
2       your    your    PRON    PRP$    Case=Gen|Person=2|Poss=Yes|PronType=Prs 4       nmod:poss       4:nmod:poss     _
3       political       political       ADJ     JJ      Degree=Pos      4       amod    4:amod  _
4       persuasion      persuasion      NOUN    NN      Number=Sing     1       nsubj   1:nsubj SpaceAfter=No

whereas in this case, it is acting as a determiner?

# sent_id = newsgroup-groups.google.com_misc.consumers_a534e32067078b08_ENG_20060116_030800-0120
# text = The dangers implicit in any attack upon that nation, however, seem to significantly offset whatever gains could be made in the so-called "War on Terror."
14      to      to      PART    TO      _       16      mark    16:mark _
15      significantly   significantly   ADV     RB      _       16      advmod  16:advmod       _
16      offset  offset  VERB    VB      VerbForm=Inf    13      xcomp   13:xcomp        _
17      whatever        whatever        DET     WDT     PronType=Int    18      det     18:det  _
18      gains   gain    NOUN    NNS     Number=Plur     16      obj     16:obj  _
19      could   could   AUX     MD      VerbForm=Fin    21      aux     21:aux  _
20      be      be      AUX     VB      VerbForm=Inf    21      aux:pass        21:aux:pass     _
21      made    make    VERB    VBN     Tense=Past|VerbForm=Part|Voice=Pass     18      acl:relcl       18:acl:relcl    _

then this one seems different from the other two

# sent_id = answers-20111108104350AAp4hGP_ans-0023
# text = If you can find a pet at a shelter, or that a friend needs to get rid of because they're moving or whatever, you might be able to talk your parents into "giving it a good home" where you might not be able to talk them into buying a new pet.
20      because because SCONJ   IN      _       23      mark    23:mark _
21-22   they're _       _       _       _       _       _       _       _
21      they    they    PRON    PRP     Case=Nom|Number=Plur|Person=3|PronType=Prs      23      nsubj   23:nsubj        _
22      're     be      AUX     VBP     Mood=Ind|Number=Plur|Person=3|Tense=Pres|VerbForm=Fin   23      aux     23:aux  _
23      moving  move    VERB    VBG     Tense=Pres|VerbForm=Part        17      advcl   17:advcl:because        _
24      or      or      CCONJ   CC      _       25      cc      25:cc   _
25      whatever        whatever        DET     WDT     PronType=Int    9       conj    4:obl:at|9:conj:or      SpaceAfter=No
amir-zeldes commented 1 year ago

An example of the error you have in mind is this, where it's the head of the noun phrase?

Yes, of the three conllu examples in your post, I think only the middle one should be WDT, because it acts as a determiner. The others have typical head deprels.

nschneid commented 1 year ago

The WH-ever pronouns enter into some tricky constructions involving interrogative/relative clauses with particular meanings. I have not yet figured out what to do with these dependent WH cases from #278. See also the "no matter" subsection of the English RC guidelines (still TBD).

Regarding WDT vs. WP, though, the PTB guidelines are clear:

image
nschneid commented 1 year ago

And: I suspect the 1 UH token is an error.

AngledLuffa commented 1 year ago

I suspect the 1 UH token is an error.

How would you label "Whatever" as a complete sentence, then?

nschneid commented 1 year ago

WP. I am guessing that "boy" in "Oh boy!" and "Dude!" would both be tagged as nouns, and "Amazing!" as an adjective, though they head an utterance with exclamatory meaning.

amir-zeldes commented 1 year ago

Fixed in GUM upstream

nschneid commented 10 months ago

WP. I am guessing that "boy" in "Oh boy!" and "Dude!" would both be tagged as nouns, and "Amazing!" as an adjective, though they head an utterance with exclamatory meaning.

Though I am keeping INTJ as the UPOS at least for now (cf. #429) because I am not sure what features would be appropriate if tagged as PRON.

nschneid commented 10 months ago

Note that the structure of "whatever" adjuncts is still up in the air (are they relative or interrogative?)—related to "no matter" (#376).