spaCy/spacy/lang/nb/morph_rules.py
Ines Montani eddeb36c96
💫 Tidy up and auto-format .py files (#2983)
<!--- Provide a general summary of your changes in the title. -->

## Description
- [x] Use [`black`](https://github.com/ambv/black) to auto-format all `.py` files.
- [x] Update flake8 config to exclude very large files (lemmatization tables etc.)
- [x] Update code to be compatible with flake8 rules
- [x] Fix various small bugs, inconsistencies and messy stuff in the language data
- [x] Update docs to explain new code style (`black`, `flake8`, when to use `# fmt: off` and `# fmt: on` and what `# noqa` means)

Once #2932 is merged, which auto-formats and tidies up the CLI, we'll be able to run `flake8 spacy` actually get meaningful results.

At the moment, the code style and linting isn't applied automatically, but I'm hoping that the new [GitHub Actions](https://github.com/features/actions) will let us auto-format pull requests and post comments with relevant linting information.

### Types of change
enhancement, code style

## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [x] My changes don't require a change to the documentation, or if they do, I've added all required information.
2018-11-30 17:03:03 +01:00

669 lines
21 KiB
Python

# encoding: utf8
from __future__ import unicode_literals
from ...symbols import LEMMA, PRON_LEMMA
# This dict includes all the PRON and DET tag combinations found in the
# dataset developed by Schibsted, Nasjonalbiblioteket and LTG (to be published
# autumn 2018) and the rarely used polite form.
MORPH_RULES = {
"PRON__Animacy=Anim|Case=Nom|Number=Sing|Person=1|PronType=Prs": {
"jeg": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "One",
"Number": "Sing",
"Case": "Nom",
}
},
"PRON__Animacy=Anim|Case=Nom|Number=Sing|Person=2|PronType=Prs": {
"du": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Sing",
"Case": "Nom",
},
# polite form, not sure about the tag
"De": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Sing",
"Case": "Nom",
"Polite": "Form",
},
},
"PRON__Animacy=Anim|Case=Nom|Gender=Fem|Number=Sing|Person=3|PronType=Prs": {
"hun": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Fem",
"Case": "Nom",
}
},
"PRON__Animacy=Anim|Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs": {
"han": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Masc",
"Case": "Nom",
}
},
"PRON__Gender=Neut|Number=Sing|Person=3|PronType=Prs": {
"det": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Neut",
},
"alt": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Neut",
},
"intet": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Neut",
},
"noe": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Sing",
"Person": "Three",
"Gender": "Neut",
},
},
"PRON__Animacy=Anim|Case=Nom|Number=Plur|Person=1|PronType=Prs": {
"vi": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "One",
"Number": "Plur",
"Case": "Nom",
}
},
"PRON__Animacy=Anim|Case=Nom|Number=Plur|Person=2|PronType=Prs": {
"dere": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Plur",
"Case": "Nom",
}
},
"PRON__Case=Nom|Number=Plur|Person=3|PronType=Prs": {
"de": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Plur",
"Case": "Nom",
}
},
"PRON__Animacy=Anim|Case=Acc|Number=Sing|Person=1|PronType=Prs": {
"meg": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "One",
"Number": "Sing",
"Case": "Acc",
}
},
"PRON__Animacy=Anim|Case=Acc|Number=Sing|Person=2|PronType=Prs": {
"deg": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Sing",
"Case": "Acc",
},
# polite form, not sure about the tag
"Dem": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Sing",
"Case": "Acc",
"Polite": "Form",
},
},
"PRON__Animacy=Anim|Case=Acc|Gender=Fem|Number=Sing|Person=3|PronType=Prs": {
"henne": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Fem",
"Case": "Acc",
}
},
"PRON__Animacy=Anim|Case=Acc|Gender=Masc|Number=Sing|Person=3|PronType=Prs": {
"ham": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Masc",
"Case": "Acc",
},
"han": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Sing",
"Gender": "Masc",
"Case": "Acc",
},
},
"PRON__Animacy=Anim|Case=Acc|Number=Plur|Person=1|PronType=Prs": {
"oss": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "One",
"Number": "Plur",
"Case": "Acc",
}
},
"PRON__Animacy=Anim|Case=Acc|Number=Plur|Person=2|PronType=Prs": {
"dere": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Two",
"Number": "Plur",
"Case": "Acc",
}
},
"PRON__Case=Acc|Number=Plur|Person=3|PronType=Prs": {
"dem": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Person": "Three",
"Number": "Plur",
"Case": "Acc",
}
},
"PRON__Case=Acc|Reflex=Yes": {
"seg": {
LEMMA: PRON_LEMMA,
"Person": "Three",
"Number": ("Sing", "Plur"),
"Reflex": "Yes",
}
},
"PRON__Animacy=Anim|Case=Nom|Number=Sing|PronType=Prs": {
"man": {LEMMA: PRON_LEMMA, "PronType": "Prs", "Number": "Sing", "Case": "Nom"}
},
"DET__Gender=Masc|Number=Sing|Poss=Yes": {
"min": {
LEMMA: "min",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
"din": {
LEMMA: "din",
"Person": "Two",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
"hennes": {
LEMMA: "hennes",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
"hans": {
LEMMA: "hans",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
"sin": {
LEMMA: "sin",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
"Reflex": "Yes",
},
"vår": {
LEMMA: "vår",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
"deres": {
LEMMA: "deres",
"Person": ("Two", "Three"),
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
},
# polite form, not sure about the tag
"Deres": {
LEMMA: "Deres",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Masc",
"Polite": "Form",
},
},
"DET__Gender=Fem|Number=Sing|Poss=Yes": {
"mi": {
LEMMA: "min",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
"di": {
LEMMA: "din",
"Person": "Two",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
"hennes": {
LEMMA: "hennes",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
"hans": {
LEMMA: "hans",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
"si": {
LEMMA: "sin",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
"Reflex": "Yes",
},
"vår": {
LEMMA: "vår",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
"deres": {
LEMMA: "deres",
"Person": ("Two", "Three"),
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
},
# polite form, not sure about the tag
"Deres": {
LEMMA: "Deres",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Fem",
"Polite": "Form",
},
},
"DET__Gender=Neut|Number=Sing|Poss=Yes": {
"mitt": {
LEMMA: "min",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
"ditt": {
LEMMA: "din",
"Person": "Two",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
"hennes": {
LEMMA: "hennes",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
"hans": {
LEMMA: "hans",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
"sitt": {
LEMMA: "sin",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
"Reflex": "Yes",
},
"vårt": {
LEMMA: "vår",
"Person": "One",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
"deres": {
LEMMA: "deres",
"Person": ("Two", "Three"),
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
},
# polite form, not sure about the tag
"Deres": {
LEMMA: "Deres",
"Person": "Three",
"Number": "Sing",
"Poss": "Yes",
"Gender": "Neut",
"Polite": "Form",
},
},
"DET__Number=Plur|Poss=Yes": {
"mine": {LEMMA: "min", "Person": "One", "Number": "Plur", "Poss": "Yes"},
"dine": {LEMMA: "din", "Person": "Two", "Number": "Plur", "Poss": "Yes"},
"hennes": {LEMMA: "hennes", "Person": "Three", "Number": "Plur", "Poss": "Yes"},
"hans": {LEMMA: "hans", "Person": "Three", "Number": "Plur", "Poss": "Yes"},
"sine": {
LEMMA: "sin",
"Person": "Three",
"Number": "Plur",
"Poss": "Yes",
"Reflex": "Yes",
},
"våre": {LEMMA: "vår", "Person": "One", "Number": "Plur", "Poss": "Yes"},
"deres": {
LEMMA: "deres",
"Person": ("Two", "Three"),
"Number": "Plur",
"Poss": "Yes",
},
},
"PRON__Animacy=Anim|Number=Plur|PronType=Rcp": {
"hverandre": {LEMMA: PRON_LEMMA, "PronType": "Rcp", "Number": "Plur"}
},
"DET__Number=Plur|Poss=Yes|PronType=Rcp": {
"hverandres": {
LEMMA: "hverandres",
"PronType": "Rcp",
"Number": "Plur",
"Poss": "Yes",
}
},
"PRON___": {"som": {LEMMA: PRON_LEMMA}, "ikkenoe": {LEMMA: PRON_LEMMA}},
"PRON__PronType=Int": {"hva": {LEMMA: PRON_LEMMA, "PronType": "Int"}},
"PRON__Animacy=Anim|PronType=Int": {"hvem": {LEMMA: PRON_LEMMA, "PronType": "Int"}},
"PRON__Animacy=Anim|Poss=Yes|PronType=Int": {
"hvis": {LEMMA: PRON_LEMMA, "PronType": "Int", "Poss": "Yes"}
},
"PRON__Number=Plur|Person=3|PronType=Prs": {
"noen": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Plur",
"Person": "Three",
},
"ingen": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Plur",
"Person": "Three",
},
"alle": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Plur",
"Person": "Three",
},
},
"PRON__Gender=Fem,Masc|Number=Sing|Person=3|PronType=Prs": {
"noen": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Sing",
"Person": "Three",
"Gender": ("Fem", "Masc"),
},
"den": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Sing",
"Person": "Three",
"Gender": ("Fem", "Masc"),
},
"ingen": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Sing",
"Person": "Three",
"Gender": ("Fem", "Masc"),
"Polarity": "Neg",
},
},
"PRON__Number=Sing": {"ingenting": {LEMMA: PRON_LEMMA, "Number": "Sing"}},
"PRON__Animacy=Anim|Number=Sing|PronType=Prs": {
"en": {LEMMA: PRON_LEMMA, "PronType": "Prs", "Number": "Sing"}
},
"PRON__Animacy=Anim|Case=Gen,Nom|Number=Sing|PronType=Prs": {
"ens": {
LEMMA: PRON_LEMMA,
"PronType": "Prs",
"Number": "Sing",
"Case": ("Gen", "Nom"),
}
},
"PRON__Animacy=Anim|Case=Gen|Number=Sing|PronType=Prs": {
"ens": {LEMMA: PRON_LEMMA, "PronType": "Prs", "Number": "Sing", "Case": "Gen"}
},
"DET__Case=Gen|Gender=Masc|Number=Sing": {
"ens": {LEMMA: "en", "Number": "Sing", "Case": "Gen"}
},
"DET__Gender=Masc|Number=Sing": {
"enhver": {LEMMA: "enhver", "Number": "Sing", "Gender": "Masc"},
"all": {LEMMA: "all", "Number": "Sing", "Gender": "Masc"},
"hver": {LEMMA: "hver", "Number": "Sing", "Gender": "Masc"},
"noen": {LEMMA: "noen", "Gender": "Masc", "Number": "Sing"},
"noe": {LEMMA: "noen", "Gender": "Masc", "Number": "Sing"},
"en": {LEMMA: "en", "Number": "Sing", "Gender": "Neut"},
"ingen": {LEMMA: "ingen", "Gender": "Masc", "Number": "Sing"},
},
"DET__Gender=Fem|Number=Sing": {
"enhver": {LEMMA: "enhver", "Number": "Sing", "Gender": "Fem"},
"all": {LEMMA: "all", "Number": "Sing", "Gender": "Fem"},
"hver": {LEMMA: "hver", "Number": "Sing", "Gender": "Fem"},
"noen": {LEMMA: "noen", "Gender": "Fem", "Number": "Sing"},
"noe": {LEMMA: "noen", "Gender": "Fem", "Number": "Sing"},
"ei": {LEMMA: "en", "Number": "Sing", "Gender": "Fem"},
},
"DET__Gender=Neut|Number=Sing": {
"ethvert": {LEMMA: "enhver", "Number": "Sing", "Gender": "Neut"},
"alt": {LEMMA: "all", "Number": "Sing", "Gender": "Neut"},
"hvert": {LEMMA: "hver", "Number": "Sing", "Gender": "Neut"},
"noe": {LEMMA: "noen", "Number": "Sing", "Gender": "Neut"},
"intet": {LEMMA: "ingen", "Gender": "Neut", "Number": "Sing"},
"et": {LEMMA: "en", "Number": "Sing", "Gender": "Neut"},
},
"DET__Gender=Neut|Number=Sing|PronType=Int": {
"hvilket": {
LEMMA: "hvilken",
"PronType": "Int",
"Number": "Sing",
"Gender": "Neut",
}
},
"DET__Gender=Fem|Number=Sing|PronType=Int": {
"hvilken": {
LEMMA: "hvilken",
"PronType": "Int",
"Number": "Sing",
"Gender": "Fem",
}
},
"DET__Gender=Masc|Number=Sing|PronType=Int": {
"hvilken": {
LEMMA: "hvilken",
"PronType": "Int",
"Number": "Sing",
"Gender": "Masc",
}
},
"DET__Number=Plur|PronType=Int": {
"hvilke": {LEMMA: "hvilken", "PronType": "Int", "Number": "Plur"}
},
"DET__Number=Plur": {
"alle": {LEMMA: "all", "Number": "Plur"},
"noen": {LEMMA: "noen", "Number": "Plur"},
"egne": {LEMMA: "egen", "Number": "Plur"},
"ingen": {LEMMA: "ingen", "Number": "Plur"},
},
"DET__Gender=Masc|Number=Sing|PronType=Dem": {
"den": {LEMMA: "den", "PronType": "Dem", "Number": "Sing", "Gender": "Masc"},
"slik": {LEMMA: "slik", "PronType": "Dem", "Number": "Sing", "Gender": "Masc"},
"denne": {
LEMMA: "denne",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Masc",
},
},
"DET__Gender=Fem|Number=Sing|PronType=Dem": {
"den": {LEMMA: "den", "PronType": "Dem", "Number": "Sing", "Gender": "Fem"},
"slik": {LEMMA: "slik", "PronType": "Dem", "Number": "Sing", "Gender": "Fem"},
"denne": {LEMMA: "denne", "PronType": "Dem", "Number": "Sing", "Gender": "Fem"},
},
"DET__Gender=Neut|Number=Sing|PronType=Dem": {
"det": {LEMMA: "det", "PronType": "Dem", "Number": "Sing", "Gender": "Neut"},
"slikt": {LEMMA: "slik", "PronType": "Dem", "Number": "Sing", "Gender": "Neut"},
"dette": {
LEMMA: "dette",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Neut",
},
},
"DET__Number=Plur|PronType=Dem": {
"disse": {LEMMA: "disse", "PronType": "Dem", "Number": "Plur"},
"andre": {LEMMA: "annen", "PronType": "Dem", "Number": "Plur"},
"de": {LEMMA: "de", "PronType": "Dem", "Number": "Plur"},
"slike": {LEMMA: "slik", "PronType": "Dem", "Number": "Plur"},
},
"DET__Definite=Ind|Gender=Masc|Number=Sing|PronType=Dem": {
"annen": {LEMMA: "annen", "PronType": "Dem", "Number": "Sing", "Gender": "Masc"}
},
"DET__Definite=Ind|Gender=Fem|Number=Sing|PronType=Dem": {
"annen": {LEMMA: "annen", "PronType": "Dem", "Number": "Sing", "Gender": "Fem"}
},
"DET__Definite=Ind|Gender=Neut|Number=Sing|PronType=Dem": {
"annet": {LEMMA: "annen", "PronType": "Dem", "Number": "Sing", "Gender": "Neut"}
},
"DET__Case=Gen|Definite=Ind|Gender=Masc|Number=Sing|PronType=Dem": {
"annens": {
LEMMA: "annnen",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Masc",
"Case": "Gen",
}
},
"DET__Case=Gen|Number=Plur|PronType=Dem": {
"andres": {LEMMA: "annen", "PronType": "Dem", "Number": "Plur", "Case": "Gen"}
},
"DET__Case=Gen|Gender=Fem|Number=Sing|PronType=Dem": {
"dens": {
LEMMA: "den",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Fem",
"Case": "Gen",
}
},
"DET__Case=Gen|Gender=Masc|Number=Sing|PronType=Dem": {
"hvis": {
LEMMA: "hvis",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Masc",
"Case": "Gen",
},
"dens": {
LEMMA: "den",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Masc",
"Case": "Gen",
},
},
"DET__Case=Gen|Gender=Neut|Number=Sing|PronType=Dem": {
"dets": {
LEMMA: "det",
"PronType": "Dem",
"Number": "Sing",
"Gender": "Neut",
"Case": "Gen",
}
},
"DET__Case=Gen|Number=Plur": {
"alles": {LEMMA: "all", "Number": "Plur", "Case": "Gen"}
},
"DET__Definite=Def|Number=Sing|PronType=Dem": {
"andre": {LEMMA: "annen", "Number": "Sing", "PronType": "Dem"}
},
"DET__Definite=Def|PronType=Dem": {
"samme": {LEMMA: "samme", "PronType": "Dem"},
"forrige": {LEMMA: "forrige", "PronType": "Dem"},
"neste": {LEMMA: "neste", "PronType": "Dem"},
},
"DET__Definite=Def": {"selve": {LEMMA: "selve"}, "selveste": {LEMMA: "selveste"}},
"DET___": {"selv": {LEMMA: "selv"}, "endel": {LEMMA: "endel"}},
"DET__Definite=Ind|Gender=Fem|Number=Sing": {
"egen": {LEMMA: "egen", "Gender": "Fem", "Number": "Sing"}
},
"DET__Definite=Ind|Gender=Masc|Number=Sing": {
"egen": {LEMMA: "egen", "Gender": "Masc", "Number": "Sing"}
},
"DET__Definite=Ind|Gender=Neut|Number=Sing": {
"eget": {LEMMA: "egen", "Gender": "Neut", "Number": "Sing"}
},
# same wordform and pos (verb), have to specify the exact features in order to not mix them up
"VERB__Mood=Ind|Tense=Pres|VerbForm=Fin": {
"": {LEMMA: "", "VerbForm": "Fin", "Tense": "Pres", "Mood": "Ind"}
},
"VERB__Mood=Ind|Tense=Past|VerbForm=Fin": {
"": {LEMMA: "se", "VerbForm": "Fin", "Tense": "Past", "Mood": "Ind"}
},
}
# copied from the English morph_rules.py
for tag, rules in MORPH_RULES.items():
for key, attrs in dict(rules).items():
rules[key.title()] = attrs